npm - local-llm-rn - Versions diffs - 1.0.0 - Mend

local-llm-rn 1.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (626) hide show

package/cpp/CMakeLists.txt +285 -0
package/cpp/common/CMakeLists.txt +149 -0
package/cpp/common/arg.cpp +3799 -0
package/cpp/common/arg.h +131 -0
package/cpp/common/base64.hpp +392 -0
package/cpp/common/build-info.cpp.in +4 -0
package/cpp/common/chat-parser-xml-toolcall.cpp +879 -0
package/cpp/common/chat-parser-xml-toolcall.h +45 -0
package/cpp/common/chat-parser.cpp +1649 -0
package/cpp/common/chat-parser.h +133 -0
package/cpp/common/chat-peg-parser.cpp +124 -0
package/cpp/common/chat-peg-parser.h +105 -0
package/cpp/common/chat.cpp +3355 -0
package/cpp/common/chat.h +252 -0
package/cpp/common/common.cpp +1824 -0
package/cpp/common/common.h +930 -0
package/cpp/common/console.cpp +1137 -0
package/cpp/common/console.h +41 -0
package/cpp/common/debug.cpp +167 -0
package/cpp/common/debug.h +43 -0
package/cpp/common/download.cpp +792 -0
package/cpp/common/download.h +84 -0
package/cpp/common/http.h +84 -0
package/cpp/common/jinja/README.md +88 -0
package/cpp/common/jinja/caps.cpp +285 -0
package/cpp/common/jinja/caps.h +30 -0
package/cpp/common/jinja/lexer.cpp +341 -0
package/cpp/common/jinja/lexer.h +157 -0
package/cpp/common/jinja/parser.cpp +591 -0
package/cpp/common/jinja/parser.h +21 -0
package/cpp/common/jinja/runtime.cpp +867 -0
package/cpp/common/jinja/runtime.h +638 -0
package/cpp/common/jinja/string.cpp +213 -0
package/cpp/common/jinja/string.h +61 -0
package/cpp/common/jinja/utils.h +149 -0
package/cpp/common/jinja/value.cpp +1393 -0
package/cpp/common/jinja/value.h +756 -0
package/cpp/common/json-partial.cpp +324 -0
package/cpp/common/json-partial.h +39 -0
package/cpp/common/json-schema-to-grammar.cpp +1153 -0
package/cpp/common/json-schema-to-grammar.h +43 -0
package/cpp/common/llguidance.cpp +258 -0
package/cpp/common/log.cpp +446 -0
package/cpp/common/log.h +119 -0
package/cpp/common/ngram-cache.cpp +285 -0
package/cpp/common/ngram-cache.h +101 -0
package/cpp/common/ngram-map.cpp +530 -0
package/cpp/common/ngram-map.h +115 -0
package/cpp/common/ngram-mod.cpp +60 -0
package/cpp/common/ngram-mod.h +38 -0
package/cpp/common/peg-parser.cpp +1712 -0
package/cpp/common/peg-parser.h +459 -0
package/cpp/common/preset.cpp +483 -0
package/cpp/common/preset.h +83 -0
package/cpp/common/regex-partial.cpp +204 -0
package/cpp/common/regex-partial.h +56 -0
package/cpp/common/sampling.cpp +745 -0
package/cpp/common/sampling.h +119 -0
package/cpp/common/speculative.cpp +1074 -0
package/cpp/common/speculative.h +41 -0
package/cpp/common/unicode.cpp +64 -0
package/cpp/common/unicode.h +22 -0
package/cpp/ggml/CMakeLists.txt +494 -0
package/cpp/ggml/cmake/GitVars.cmake +22 -0
package/cpp/ggml/cmake/common.cmake +50 -0
package/cpp/ggml/cmake/ggml-config.cmake.in +191 -0
package/cpp/ggml/include/ggml-alloc.h +85 -0
package/cpp/ggml/include/ggml-backend.h +373 -0
package/cpp/ggml/include/ggml-blas.h +25 -0
package/cpp/ggml/include/ggml-cann.h +123 -0
package/cpp/ggml/include/ggml-cpp.h +39 -0
package/cpp/ggml/include/ggml-cpu.h +151 -0
package/cpp/ggml/include/ggml-cuda.h +47 -0
package/cpp/ggml/include/ggml-hexagon.h +19 -0
package/cpp/ggml/include/ggml-metal.h +61 -0
package/cpp/ggml/include/ggml-opencl.h +26 -0
package/cpp/ggml/include/ggml-opt.h +256 -0
package/cpp/ggml/include/ggml-rpc.h +30 -0
package/cpp/ggml/include/ggml-sycl.h +49 -0
package/cpp/ggml/include/ggml-virtgpu.h +14 -0
package/cpp/ggml/include/ggml-vulkan.h +29 -0
package/cpp/ggml/include/ggml-webgpu.h +19 -0
package/cpp/ggml/include/ggml-zdnn.h +17 -0
package/cpp/ggml/include/ggml-zendnn.h +22 -0
package/cpp/ggml/include/ggml.h +2753 -0
package/cpp/ggml/include/gguf.h +204 -0
package/cpp/ggml/src/CMakeLists.txt +492 -0
package/cpp/ggml/src/ggml-alloc.c +1244 -0
package/cpp/ggml/src/ggml-backend-dl.cpp +48 -0
package/cpp/ggml/src/ggml-backend-dl.h +45 -0
package/cpp/ggml/src/ggml-backend-impl.h +255 -0
package/cpp/ggml/src/ggml-backend-reg.cpp +566 -0
package/cpp/ggml/src/ggml-backend.cpp +2270 -0
package/cpp/ggml/src/ggml-blas/CMakeLists.txt +101 -0
package/cpp/ggml/src/ggml-blas/ggml-blas.cpp +518 -0
package/cpp/ggml/src/ggml-common.h +1878 -0
package/cpp/ggml/src/ggml-cpu/CMakeLists.txt +691 -0
package/cpp/ggml/src/ggml-cpu/amx/amx.cpp +247 -0
package/cpp/ggml/src/ggml-cpu/amx/amx.h +8 -0
package/cpp/ggml/src/ggml-cpu/amx/common.h +91 -0
package/cpp/ggml/src/ggml-cpu/amx/mmq.cpp +2512 -0
package/cpp/ggml/src/ggml-cpu/amx/mmq.h +10 -0
package/cpp/ggml/src/ggml-cpu/arch/arm/cpu-feats.cpp +98 -0
package/cpp/ggml/src/ggml-cpu/arch/arm/quants.c +4052 -0
package/cpp/ggml/src/ggml-cpu/arch/arm/repack.cpp +4935 -0
package/cpp/ggml/src/ggml-cpu/arch/loongarch/quants.c +2159 -0
package/cpp/ggml/src/ggml-cpu/arch/powerpc/cpu-feats.cpp +82 -0
package/cpp/ggml/src/ggml-cpu/arch/powerpc/quants.c +2305 -0
package/cpp/ggml/src/ggml-cpu/arch/riscv/cpu-feats.cpp +38 -0
package/cpp/ggml/src/ggml-cpu/arch/riscv/quants.c +2726 -0
package/cpp/ggml/src/ggml-cpu/arch/riscv/repack.cpp +342 -0
package/cpp/ggml/src/ggml-cpu/arch/s390/cpu-feats.cpp +50 -0
package/cpp/ggml/src/ggml-cpu/arch/s390/quants.c +1468 -0
package/cpp/ggml/src/ggml-cpu/arch/wasm/quants.c +1221 -0
package/cpp/ggml/src/ggml-cpu/arch/x86/cpu-feats.cpp +327 -0
package/cpp/ggml/src/ggml-cpu/arch/x86/quants.c +3820 -0
package/cpp/ggml/src/ggml-cpu/arch/x86/repack.cpp +6307 -0
package/cpp/ggml/src/ggml-cpu/arch-fallback.h +313 -0
package/cpp/ggml/src/ggml-cpu/binary-ops.cpp +154 -0
package/cpp/ggml/src/ggml-cpu/binary-ops.h +16 -0
package/cpp/ggml/src/ggml-cpu/cmake/FindSIMD.cmake +100 -0
package/cpp/ggml/src/ggml-cpu/common.h +95 -0
package/cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h +529 -0
package/cpp/ggml/src/ggml-cpu/ggml-cpu.c +3734 -0
package/cpp/ggml/src/ggml-cpu/ggml-cpu.cpp +701 -0
package/cpp/ggml/src/ggml-cpu/hbm.cpp +55 -0
package/cpp/ggml/src/ggml-cpu/hbm.h +8 -0
package/cpp/ggml/src/ggml-cpu/kleidiai/kernels.cpp +938 -0
package/cpp/ggml/src/ggml-cpu/kleidiai/kernels.h +90 -0
package/cpp/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +798 -0
package/cpp/ggml/src/ggml-cpu/kleidiai/kleidiai.h +17 -0
package/cpp/ggml/src/ggml-cpu/llamafile/sgemm.cpp +4033 -0
package/cpp/ggml/src/ggml-cpu/llamafile/sgemm.h +25 -0
package/cpp/ggml/src/ggml-cpu/ops.cpp +10978 -0
package/cpp/ggml/src/ggml-cpu/ops.h +116 -0
package/cpp/ggml/src/ggml-cpu/quants.c +1193 -0
package/cpp/ggml/src/ggml-cpu/quants.h +97 -0
package/cpp/ggml/src/ggml-cpu/repack.cpp +3316 -0
package/cpp/ggml/src/ggml-cpu/repack.h +173 -0
package/cpp/ggml/src/ggml-cpu/simd-gemm.h +136 -0
package/cpp/ggml/src/ggml-cpu/simd-mappings.h +1279 -0
package/cpp/ggml/src/ggml-cpu/spacemit/ime.cpp +1025 -0
package/cpp/ggml/src/ggml-cpu/spacemit/ime.h +13 -0
package/cpp/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +3196 -0
package/cpp/ggml/src/ggml-cpu/spacemit/ime_kernels.h +26 -0
package/cpp/ggml/src/ggml-cpu/traits.cpp +36 -0
package/cpp/ggml/src/ggml-cpu/traits.h +38 -0
package/cpp/ggml/src/ggml-cpu/unary-ops.cpp +337 -0
package/cpp/ggml/src/ggml-cpu/unary-ops.h +35 -0
package/cpp/ggml/src/ggml-cpu/vec.cpp +629 -0
package/cpp/ggml/src/ggml-cpu/vec.h +1585 -0
package/cpp/ggml/src/ggml-hexagon/CMakeLists.txt +117 -0
package/cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp +3232 -0
package/cpp/ggml/src/ggml-hexagon/htp/CMakeLists.txt +45 -0
package/cpp/ggml/src/ggml-hexagon/htp/act-ops.c +815 -0
package/cpp/ggml/src/ggml-hexagon/htp/argsort-ops.c +281 -0
package/cpp/ggml/src/ggml-hexagon/htp/binary-ops.c +827 -0
package/cpp/ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake +157 -0
package/cpp/ggml/src/ggml-hexagon/htp/cpy-ops.c +251 -0
package/cpp/ggml/src/ggml-hexagon/htp/flash-attn-ops.c +666 -0
package/cpp/ggml/src/ggml-hexagon/htp/get-rows-ops.c +111 -0
package/cpp/ggml/src/ggml-hexagon/htp/hex-dma.c +63 -0
package/cpp/ggml/src/ggml-hexagon/htp/hex-dma.h +182 -0
package/cpp/ggml/src/ggml-hexagon/htp/hex-dump.h +77 -0
package/cpp/ggml/src/ggml-hexagon/htp/hex-fastdiv.h +37 -0
package/cpp/ggml/src/ggml-hexagon/htp/hex-utils.h +51 -0
package/cpp/ggml/src/ggml-hexagon/htp/htp-ctx.h +35 -0
package/cpp/ggml/src/ggml-hexagon/htp/htp-msg.h +154 -0
package/cpp/ggml/src/ggml-hexagon/htp/htp-ops.h +65 -0
package/cpp/ggml/src/ggml-hexagon/htp/htp_iface.idl +16 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-arith.h +470 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-base.h +173 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-copy.h +245 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-div.h +116 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-dump.h +129 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-exp.h +215 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-floor.h +100 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-inverse.h +176 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-reduce.h +266 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-scale.h +133 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-sigmoid.h +141 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-sqrt.h +126 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-types.h +36 -0
package/cpp/ggml/src/ggml-hexagon/htp/hvx-utils.h +18 -0
package/cpp/ggml/src/ggml-hexagon/htp/main.c +1150 -0
package/cpp/ggml/src/ggml-hexagon/htp/matmul-ops.c +2595 -0
package/cpp/ggml/src/ggml-hexagon/htp/rope-ops.c +498 -0
package/cpp/ggml/src/ggml-hexagon/htp/set-rows-ops.c +167 -0
package/cpp/ggml/src/ggml-hexagon/htp/softmax-ops.c +421 -0
package/cpp/ggml/src/ggml-hexagon/htp/sum-rows-ops.c +130 -0
package/cpp/ggml/src/ggml-hexagon/htp/unary-ops.c +384 -0
package/cpp/ggml/src/ggml-hexagon/htp/worker-pool.c +293 -0
package/cpp/ggml/src/ggml-hexagon/htp/worker-pool.h +57 -0
package/cpp/ggml/src/ggml-hexagon/htp-drv.cpp +418 -0
package/cpp/ggml/src/ggml-hexagon/htp-drv.h +121 -0
package/cpp/ggml/src/ggml-hexagon/libdl.h +79 -0
package/cpp/ggml/src/ggml-hexagon/libggml-htp.inf +38 -0
package/cpp/ggml/src/ggml-hexagon/op-desc.h +153 -0
package/cpp/ggml/src/ggml-impl.h +724 -0
package/cpp/ggml/src/ggml-metal/CMakeLists.txt +124 -0
package/cpp/ggml/src/ggml-metal/ggml-metal-common.cpp +457 -0
package/cpp/ggml/src/ggml-metal/ggml-metal-common.h +52 -0
package/cpp/ggml/src/ggml-metal/ggml-metal-context.h +41 -0
package/cpp/ggml/src/ggml-metal/ggml-metal-context.m +702 -0
package/cpp/ggml/src/ggml-metal/ggml-metal-device.cpp +1890 -0
package/cpp/ggml/src/ggml-metal/ggml-metal-device.h +290 -0
package/cpp/ggml/src/ggml-metal/ggml-metal-device.m +1749 -0
package/cpp/ggml/src/ggml-metal/ggml-metal-impl.h +1054 -0
package/cpp/ggml/src/ggml-metal/ggml-metal-ops.cpp +4370 -0
package/cpp/ggml/src/ggml-metal/ggml-metal-ops.h +94 -0
package/cpp/ggml/src/ggml-metal/ggml-metal.cpp +937 -0
package/cpp/ggml/src/ggml-metal/ggml-metal.metal +9819 -0
package/cpp/ggml/src/ggml-musa/CMakeLists.txt +125 -0
package/cpp/ggml/src/ggml-musa/mudnn.cu +112 -0
package/cpp/ggml/src/ggml-musa/mudnn.cuh +12 -0
package/cpp/ggml/src/ggml-opencl/CMakeLists.txt +150 -0
package/cpp/ggml/src/ggml-opencl/ggml-opencl.cpp +11553 -0
package/cpp/ggml/src/ggml-opencl/kernels/add.cl +190 -0
package/cpp/ggml/src/ggml-opencl/kernels/add_id.cl +42 -0
package/cpp/ggml/src/ggml-opencl/kernels/argsort.cl +86 -0
package/cpp/ggml/src/ggml-opencl/kernels/clamp.cl +20 -0
package/cpp/ggml/src/ggml-opencl/kernels/concat.cl +51 -0
package/cpp/ggml/src/ggml-opencl/kernels/conv2d.cl +185 -0
package/cpp/ggml/src/ggml-opencl/kernels/conv2d_f16_f32.cl +176 -0
package/cpp/ggml/src/ggml-opencl/kernels/cpy.cl +184 -0
package/cpp/ggml/src/ggml-opencl/kernels/cvt.cl +417 -0
package/cpp/ggml/src/ggml-opencl/kernels/diag_mask_inf.cl +58 -0
package/cpp/ggml/src/ggml-opencl/kernels/div.cl +138 -0
package/cpp/ggml/src/ggml-opencl/kernels/embed_kernel.py +26 -0
package/cpp/ggml/src/ggml-opencl/kernels/expm1.cl +113 -0
package/cpp/ggml/src/ggml-opencl/kernels/fill.cl +17 -0
package/cpp/ggml/src/ggml-opencl/kernels/flash_attn_f16.cl +370 -0
package/cpp/ggml/src/ggml-opencl/kernels/flash_attn_f32.cl +371 -0
package/cpp/ggml/src/ggml-opencl/kernels/flash_attn_f32_f16.cl +373 -0
package/cpp/ggml/src/ggml-opencl/kernels/gelu.cl +89 -0
package/cpp/ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32.cl +162 -0
package/cpp/ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32.cl +156 -0
package/cpp/ggml/src/ggml-opencl/kernels/gemv_noshuffle.cl +268 -0
package/cpp/ggml/src/ggml-opencl/kernels/gemv_noshuffle_general.cl +274 -0
package/cpp/ggml/src/ggml-opencl/kernels/gemv_noshuffle_general_q8_0_f32.cl +195 -0
package/cpp/ggml/src/ggml-opencl/kernels/get_rows.cl +187 -0
package/cpp/ggml/src/ggml-opencl/kernels/glu.cl +378 -0
package/cpp/ggml/src/ggml-opencl/kernels/group_norm.cl +121 -0
package/cpp/ggml/src/ggml-opencl/kernels/im2col_f16.cl +57 -0
package/cpp/ggml/src/ggml-opencl/kernels/im2col_f32.cl +57 -0
package/cpp/ggml/src/ggml-opencl/kernels/mean.cl +140 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul.cl +152 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mat_Ab_Bi_8x4.cl +139 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mat_f16_f32.cl +130 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_kq_kqv.cl +273 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl +146 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl +147 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mm_q4_0_f32_l4_lm.cl +163 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mm_q4_1_f32_l4_lm.cl +165 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mm_q6_k_f32_l4_lm.cl +158 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mm_q8_0_f32_8x4.cl +129 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl +154 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_f16_f16.cl +118 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_f16_f32.cl +118 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_f16_f32_1row.cl +94 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_f16_f32_l4.cl +84 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_f32_f32.cl +118 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32.cl +189 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32_flat.cl +176 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_q4_0_f32_8x_flat.cl +283 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32.cl +140 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32_flat.cl +222 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32.cl +144 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32_flat.cl +167 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32.cl +192 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_1d_16x_flat.cl +307 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_1d_8x_flat.cl +265 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_8x_flat.cl +272 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_v.cl +254 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q4_1_f32.cl +219 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q4_1_f32_flat.cl +229 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q4_k_f32.cl +180 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q6_k_f32.cl +194 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q6_k_f32_flat.cl +194 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32.cl +125 -0
package/cpp/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32_flat.cl +202 -0
package/cpp/ggml/src/ggml-opencl/kernels/norm.cl +161 -0
package/cpp/ggml/src/ggml-opencl/kernels/pad.cl +39 -0
package/cpp/ggml/src/ggml-opencl/kernels/relu.cl +16 -0
package/cpp/ggml/src/ggml-opencl/kernels/repeat.cl +38 -0
package/cpp/ggml/src/ggml-opencl/kernels/rms_norm.cl +190 -0
package/cpp/ggml/src/ggml-opencl/kernels/rope.cl +747 -0
package/cpp/ggml/src/ggml-opencl/kernels/scale.cl +27 -0
package/cpp/ggml/src/ggml-opencl/kernels/set_rows.cl +208 -0
package/cpp/ggml/src/ggml-opencl/kernels/sigmoid.cl +29 -0
package/cpp/ggml/src/ggml-opencl/kernels/silu.cl +30 -0
package/cpp/ggml/src/ggml-opencl/kernels/softmax_4_f16.cl +108 -0
package/cpp/ggml/src/ggml-opencl/kernels/softmax_4_f32.cl +108 -0
package/cpp/ggml/src/ggml-opencl/kernels/softmax_f16.cl +107 -0
package/cpp/ggml/src/ggml-opencl/kernels/softmax_f32.cl +107 -0
package/cpp/ggml/src/ggml-opencl/kernels/softplus.cl +116 -0
package/cpp/ggml/src/ggml-opencl/kernels/solve_tri.cl +51 -0
package/cpp/ggml/src/ggml-opencl/kernels/sqr.cl +53 -0
package/cpp/ggml/src/ggml-opencl/kernels/sqrt.cl +53 -0
package/cpp/ggml/src/ggml-opencl/kernels/ssm_conv.cl +77 -0
package/cpp/ggml/src/ggml-opencl/kernels/sub.cl +138 -0
package/cpp/ggml/src/ggml-opencl/kernels/sum_rows.cl +140 -0
package/cpp/ggml/src/ggml-opencl/kernels/tanh.cl +109 -0
package/cpp/ggml/src/ggml-opencl/kernels/transpose.cl +117 -0
package/cpp/ggml/src/ggml-opencl/kernels/tri.cl +32 -0
package/cpp/ggml/src/ggml-opencl/kernels/tsembd.cl +48 -0
package/cpp/ggml/src/ggml-opencl/kernels/upscale.cl +120 -0
package/cpp/ggml/src/ggml-opt.cpp +1093 -0
package/cpp/ggml/src/ggml-quants.c +5325 -0
package/cpp/ggml/src/ggml-quants.h +106 -0
package/cpp/ggml/src/ggml-rpc/CMakeLists.txt +9 -0
package/cpp/ggml/src/ggml-rpc/ggml-rpc.cpp +2118 -0
package/cpp/ggml/src/ggml-threading.cpp +12 -0
package/cpp/ggml/src/ggml-threading.h +14 -0
package/cpp/ggml/src/ggml-virtgpu/CMakeLists.txt +70 -0
package/cpp/ggml/src/ggml-virtgpu/apir_cs_ggml-rpc-front.cpp +87 -0
package/cpp/ggml/src/ggml-virtgpu/backend/CMakeLists.txt +21 -0
package/cpp/ggml/src/ggml-virtgpu/backend/apir_cs_ggml-rpc-back.cpp +115 -0
package/cpp/ggml/src/ggml-virtgpu/backend/backend-convert.h +13 -0
package/cpp/ggml/src/ggml-virtgpu/backend/backend-dispatched-backend.cpp +102 -0
package/cpp/ggml/src/ggml-virtgpu/backend/backend-dispatched-buffer-type.cpp +105 -0
package/cpp/ggml/src/ggml-virtgpu/backend/backend-dispatched-buffer.cpp +179 -0
package/cpp/ggml/src/ggml-virtgpu/backend/backend-dispatched-device.cpp +148 -0
package/cpp/ggml/src/ggml-virtgpu/backend/backend-dispatched.cpp +51 -0
package/cpp/ggml/src/ggml-virtgpu/backend/backend-dispatched.gen.h +73 -0
package/cpp/ggml/src/ggml-virtgpu/backend/backend-dispatched.h +27 -0
package/cpp/ggml/src/ggml-virtgpu/backend/backend-virgl-apir.h +32 -0
package/cpp/ggml/src/ggml-virtgpu/backend/backend.cpp +144 -0
package/cpp/ggml/src/ggml-virtgpu/backend/shared/api_remoting.h +95 -0
package/cpp/ggml/src/ggml-virtgpu/backend/shared/apir_backend.gen.h +94 -0
package/cpp/ggml/src/ggml-virtgpu/backend/shared/apir_backend.h +50 -0
package/cpp/ggml/src/ggml-virtgpu/backend/shared/apir_cs.h +378 -0
package/cpp/ggml/src/ggml-virtgpu/backend/shared/apir_cs_ggml.h +232 -0
package/cpp/ggml/src/ggml-virtgpu/backend/shared/apir_cs_rpc.h +58 -0
package/cpp/ggml/src/ggml-virtgpu/ggml-backend-buffer-type.cpp +81 -0
package/cpp/ggml/src/ggml-virtgpu/ggml-backend-buffer.cpp +119 -0
package/cpp/ggml/src/ggml-virtgpu/ggml-backend-device.cpp +158 -0
package/cpp/ggml/src/ggml-virtgpu/ggml-backend-reg.cpp +213 -0
package/cpp/ggml/src/ggml-virtgpu/ggml-backend.cpp +69 -0
package/cpp/ggml/src/ggml-virtgpu/ggml-remoting.h +71 -0
package/cpp/ggml/src/ggml-virtgpu/ggmlremoting_functions.yaml +166 -0
package/cpp/ggml/src/ggml-virtgpu/include/apir_hw.h +9 -0
package/cpp/ggml/src/ggml-virtgpu/regenerate_remoting.py +333 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-apir.h +15 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-forward-backend.cpp +58 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-forward-buffer-type.cpp +110 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-forward-buffer.cpp +173 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-forward-device.cpp +192 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-forward-impl.h +36 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-forward.gen.h +53 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-shm.cpp +98 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-shm.h +23 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-utils.cpp +179 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu-utils.h +86 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu.cpp +544 -0
package/cpp/ggml/src/ggml-virtgpu/virtgpu.h +117 -0
package/cpp/ggml/src/ggml-webgpu/CMakeLists.txt +80 -0
package/cpp/ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp +1231 -0
package/cpp/ggml/src/ggml-webgpu/ggml-webgpu.cpp +3150 -0
package/cpp/ggml/src/ggml-webgpu/pre_wgsl.hpp +778 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/argmax.wgsl +72 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/argsort.wgsl +106 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/argsort_merge.wgsl +134 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/binary.wgsl +107 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +923 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/cpy.tmpl.wgsl +107 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/cumsum.wgsl +66 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +182 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn.wgsl +636 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.wgsl +668 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/glu.tmpl.wgsl +323 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/memset.wgsl +40 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.wgsl +713 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl +103 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.wgsl +138 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_subgroup_matrix.wgsl +188 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.wgsl +194 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/pad.wgsl +86 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +123 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/rope.tmpl.wgsl +295 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/scale.wgsl +63 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl +109 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/soft_max.tmpl.wgsl +345 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/sum_rows.wgsl +55 -0
package/cpp/ggml/src/ggml-webgpu/wgsl-shaders/unary.wgsl +193 -0
package/cpp/ggml/src/ggml-zdnn/CMakeLists.txt +36 -0
package/cpp/ggml/src/ggml-zdnn/common.hpp +59 -0
package/cpp/ggml/src/ggml-zdnn/ggml-zdnn.cpp +633 -0
package/cpp/ggml/src/ggml-zdnn/mmf.cpp +80 -0
package/cpp/ggml/src/ggml-zdnn/mmf.hpp +12 -0
package/cpp/ggml/src/ggml-zdnn/utils.cpp +79 -0
package/cpp/ggml/src/ggml-zdnn/utils.hpp +19 -0
package/cpp/ggml/src/ggml-zendnn/CMakeLists.txt +92 -0
package/cpp/ggml/src/ggml-zendnn/ggml-zendnn.cpp +469 -0
package/cpp/ggml/src/ggml.c +7669 -0
package/cpp/ggml/src/ggml.cpp +26 -0
package/cpp/ggml/src/gguf.cpp +1699 -0
package/cpp/include/llama-cpp.h +32 -0
package/cpp/include/llama.h +1568 -0
package/cpp/mtmd/CMakeLists.txt +98 -0
package/cpp/mtmd/README.md +63 -0
package/cpp/mtmd/clip-graph.h +117 -0
package/cpp/mtmd/clip-impl.h +586 -0
package/cpp/mtmd/clip-model.h +390 -0
package/cpp/mtmd/clip.cpp +4154 -0
package/cpp/mtmd/clip.h +121 -0
package/cpp/mtmd/deprecation-warning.cpp +22 -0
package/cpp/mtmd/legacy-models/convert_image_encoder_to_gguf.py +412 -0
package/cpp/mtmd/legacy-models/glmedge-convert-image-encoder-to-gguf.py +280 -0
package/cpp/mtmd/legacy-models/glmedge-surgery.py +33 -0
package/cpp/mtmd/legacy-models/llava_surgery.py +38 -0
package/cpp/mtmd/legacy-models/llava_surgery_v2.py +180 -0
package/cpp/mtmd/legacy-models/minicpmv-convert-image-encoder-to-gguf.py +892 -0
package/cpp/mtmd/legacy-models/minicpmv-surgery.py +47 -0
package/cpp/mtmd/models/cogvlm.cpp +98 -0
package/cpp/mtmd/models/conformer.cpp +216 -0
package/cpp/mtmd/models/glm4v.cpp +122 -0
package/cpp/mtmd/models/internvl.cpp +69 -0
package/cpp/mtmd/models/kimik25.cpp +101 -0
package/cpp/mtmd/models/kimivl.cpp +63 -0
package/cpp/mtmd/models/llama4.cpp +96 -0
package/cpp/mtmd/models/llava.cpp +374 -0
package/cpp/mtmd/models/minicpmv.cpp +114 -0
package/cpp/mtmd/models/mobilenetv5.cpp +451 -0
package/cpp/mtmd/models/models.h +128 -0
package/cpp/mtmd/models/nemotron-v2-vl.cpp +35 -0
package/cpp/mtmd/models/paddleocr.cpp +52 -0
package/cpp/mtmd/models/pixtral.cpp +86 -0
package/cpp/mtmd/models/qwen2vl.cpp +183 -0
package/cpp/mtmd/models/qwen3vl.cpp +193 -0
package/cpp/mtmd/models/siglip.cpp +86 -0
package/cpp/mtmd/models/whisper-enc.cpp +115 -0
package/cpp/mtmd/models/youtuvl.cpp +179 -0
package/cpp/mtmd/mtmd-audio.cpp +730 -0
package/cpp/mtmd/mtmd-audio.h +113 -0
package/cpp/mtmd/mtmd-cli.cpp +437 -0
package/cpp/mtmd/mtmd-helper.cpp +521 -0
package/cpp/mtmd/mtmd-helper.h +96 -0
package/cpp/mtmd/mtmd.cpp +1156 -0
package/cpp/mtmd/mtmd.h +319 -0
package/cpp/mtmd/requirements.txt +5 -0
package/cpp/mtmd/test-1.jpeg +0 -0
package/cpp/mtmd/test-2.mp3 +0 -0
package/cpp/mtmd/tests.sh +192 -0
package/cpp/src/CMakeLists.txt +169 -0
package/cpp/src/llama-adapter.cpp +488 -0
package/cpp/src/llama-adapter.h +89 -0
package/cpp/src/llama-arch.cpp +2855 -0
package/cpp/src/llama-arch.h +619 -0
package/cpp/src/llama-batch.cpp +917 -0
package/cpp/src/llama-batch.h +173 -0
package/cpp/src/llama-chat.cpp +896 -0
package/cpp/src/llama-chat.h +71 -0
package/cpp/src/llama-context.cpp +3512 -0
package/cpp/src/llama-context.h +359 -0
package/cpp/src/llama-cparams.cpp +5 -0
package/cpp/src/llama-cparams.h +44 -0
package/cpp/src/llama-grammar.cpp +1464 -0
package/cpp/src/llama-grammar.h +194 -0
package/cpp/src/llama-graph.cpp +2685 -0
package/cpp/src/llama-graph.h +1026 -0
package/cpp/src/llama-hparams.cpp +234 -0
package/cpp/src/llama-hparams.h +339 -0
package/cpp/src/llama-impl.cpp +171 -0
package/cpp/src/llama-impl.h +73 -0
package/cpp/src/llama-io.cpp +15 -0
package/cpp/src/llama-io.h +35 -0
package/cpp/src/llama-kv-cache-iswa.cpp +330 -0
package/cpp/src/llama-kv-cache-iswa.h +137 -0
package/cpp/src/llama-kv-cache.cpp +2271 -0
package/cpp/src/llama-kv-cache.h +388 -0
package/cpp/src/llama-kv-cells.h +533 -0
package/cpp/src/llama-memory-hybrid-iswa.cpp +275 -0
package/cpp/src/llama-memory-hybrid-iswa.h +140 -0
package/cpp/src/llama-memory-hybrid.cpp +268 -0
package/cpp/src/llama-memory-hybrid.h +139 -0
package/cpp/src/llama-memory-recurrent.cpp +1165 -0
package/cpp/src/llama-memory-recurrent.h +182 -0
package/cpp/src/llama-memory.cpp +59 -0
package/cpp/src/llama-memory.h +122 -0
package/cpp/src/llama-mmap.cpp +785 -0
package/cpp/src/llama-mmap.h +92 -0
package/cpp/src/llama-model-loader.cpp +1414 -0
package/cpp/src/llama-model-loader.h +203 -0
package/cpp/src/llama-model-saver.cpp +286 -0
package/cpp/src/llama-model-saver.h +37 -0
package/cpp/src/llama-model.cpp +9253 -0
package/cpp/src/llama-model.h +576 -0
package/cpp/src/llama-quant.cpp +1119 -0
package/cpp/src/llama-quant.h +1 -0
package/cpp/src/llama-sampler.cpp +3885 -0
package/cpp/src/llama-sampler.h +42 -0
package/cpp/src/llama-vocab.cpp +3970 -0
package/cpp/src/llama-vocab.h +187 -0
package/cpp/src/llama.cpp +1313 -0
package/cpp/src/models/afmoe.cpp +191 -0
package/cpp/src/models/apertus.cpp +125 -0
package/cpp/src/models/arcee.cpp +135 -0
package/cpp/src/models/arctic.cpp +138 -0
package/cpp/src/models/arwkv7.cpp +86 -0
package/cpp/src/models/baichuan.cpp +122 -0
package/cpp/src/models/bailingmoe.cpp +144 -0
package/cpp/src/models/bailingmoe2.cpp +135 -0
package/cpp/src/models/bert.cpp +178 -0
package/cpp/src/models/bitnet.cpp +160 -0
package/cpp/src/models/bloom.cpp +101 -0
package/cpp/src/models/chameleon.cpp +178 -0
package/cpp/src/models/chatglm.cpp +132 -0
package/cpp/src/models/codeshell.cpp +111 -0
package/cpp/src/models/cogvlm.cpp +102 -0
package/cpp/src/models/cohere2-iswa.cpp +134 -0
package/cpp/src/models/command-r.cpp +122 -0
package/cpp/src/models/dbrx.cpp +123 -0
package/cpp/src/models/deci.cpp +135 -0
package/cpp/src/models/deepseek.cpp +144 -0
package/cpp/src/models/deepseek2.cpp +262 -0
package/cpp/src/models/delta-net-base.cpp +376 -0
package/cpp/src/models/dots1.cpp +134 -0
package/cpp/src/models/dream.cpp +105 -0
package/cpp/src/models/ernie4-5-moe.cpp +150 -0
package/cpp/src/models/ernie4-5.cpp +110 -0
package/cpp/src/models/eurobert.cpp +97 -0
package/cpp/src/models/exaone-moe.cpp +146 -0
package/cpp/src/models/exaone.cpp +114 -0
package/cpp/src/models/exaone4.cpp +123 -0
package/cpp/src/models/falcon-h1.cpp +111 -0
package/cpp/src/models/falcon.cpp +120 -0
package/cpp/src/models/gemma-embedding.cpp +116 -0
package/cpp/src/models/gemma.cpp +112 -0
package/cpp/src/models/gemma2-iswa.cpp +128 -0
package/cpp/src/models/gemma3.cpp +155 -0
package/cpp/src/models/gemma3n-iswa.cpp +384 -0
package/cpp/src/models/glm4-moe.cpp +170 -0
package/cpp/src/models/glm4.cpp +157 -0
package/cpp/src/models/gpt2.cpp +105 -0
package/cpp/src/models/gptneox.cpp +144 -0
package/cpp/src/models/granite-hybrid.cpp +196 -0
package/cpp/src/models/granite.cpp +211 -0
package/cpp/src/models/grok.cpp +159 -0
package/cpp/src/models/grovemoe.cpp +141 -0
package/cpp/src/models/hunyuan-dense.cpp +132 -0
package/cpp/src/models/hunyuan-moe.cpp +154 -0
package/cpp/src/models/internlm2.cpp +120 -0
package/cpp/src/models/jais.cpp +86 -0
package/cpp/src/models/jais2.cpp +123 -0
package/cpp/src/models/jamba.cpp +106 -0
package/cpp/src/models/kimi-linear.cpp +392 -0
package/cpp/src/models/lfm2.cpp +190 -0
package/cpp/src/models/llada-moe.cpp +122 -0
package/cpp/src/models/llada.cpp +99 -0
package/cpp/src/models/llama-iswa.cpp +178 -0
package/cpp/src/models/llama.cpp +168 -0
package/cpp/src/models/maincoder.cpp +117 -0
package/cpp/src/models/mamba-base.cpp +285 -0
package/cpp/src/models/mamba.cpp +54 -0
package/cpp/src/models/mimo2-iswa.cpp +123 -0
package/cpp/src/models/minicpm3.cpp +200 -0
package/cpp/src/models/minimax-m2.cpp +124 -0
package/cpp/src/models/mistral3.cpp +160 -0
package/cpp/src/models/models.h +684 -0
package/cpp/src/models/modern-bert.cpp +109 -0
package/cpp/src/models/mpt.cpp +126 -0
package/cpp/src/models/nemotron-h.cpp +148 -0
package/cpp/src/models/nemotron.cpp +122 -0
package/cpp/src/models/neo-bert.cpp +104 -0
package/cpp/src/models/olmo.cpp +121 -0
package/cpp/src/models/olmo2.cpp +150 -0
package/cpp/src/models/olmoe.cpp +124 -0
package/cpp/src/models/openai-moe-iswa.cpp +127 -0
package/cpp/src/models/openelm.cpp +124 -0
package/cpp/src/models/orion.cpp +123 -0
package/cpp/src/models/paddleocr.cpp +122 -0
package/cpp/src/models/pangu-embedded.cpp +121 -0
package/cpp/src/models/phi2.cpp +121 -0
package/cpp/src/models/phi3.cpp +152 -0
package/cpp/src/models/plamo.cpp +110 -0
package/cpp/src/models/plamo2.cpp +318 -0
package/cpp/src/models/plamo3.cpp +128 -0
package/cpp/src/models/plm.cpp +169 -0
package/cpp/src/models/qwen.cpp +108 -0
package/cpp/src/models/qwen2.cpp +126 -0
package/cpp/src/models/qwen2moe.cpp +151 -0
package/cpp/src/models/qwen2vl.cpp +117 -0
package/cpp/src/models/qwen3.cpp +117 -0
package/cpp/src/models/qwen35.cpp +386 -0
package/cpp/src/models/qwen35moe.cpp +420 -0
package/cpp/src/models/qwen3moe.cpp +124 -0
package/cpp/src/models/qwen3next.cpp +525 -0
package/cpp/src/models/qwen3vl-moe.cpp +140 -0
package/cpp/src/models/qwen3vl.cpp +132 -0
package/cpp/src/models/refact.cpp +94 -0
package/cpp/src/models/rnd1.cpp +126 -0
package/cpp/src/models/rwkv6-base.cpp +164 -0
package/cpp/src/models/rwkv6.cpp +94 -0
package/cpp/src/models/rwkv6qwen2.cpp +86 -0
package/cpp/src/models/rwkv7-base.cpp +137 -0
package/cpp/src/models/rwkv7.cpp +90 -0
package/cpp/src/models/seed-oss.cpp +124 -0
package/cpp/src/models/smallthinker.cpp +126 -0
package/cpp/src/models/smollm3.cpp +128 -0
package/cpp/src/models/stablelm.cpp +146 -0
package/cpp/src/models/starcoder.cpp +100 -0
package/cpp/src/models/starcoder2.cpp +121 -0
package/cpp/src/models/step35-iswa.cpp +168 -0
package/cpp/src/models/t5-dec.cpp +166 -0
package/cpp/src/models/t5-enc.cpp +96 -0
package/cpp/src/models/wavtokenizer-dec.cpp +149 -0
package/cpp/src/models/xverse.cpp +108 -0
package/cpp/src/unicode-data.cpp +7034 -0
package/cpp/src/unicode-data.h +20 -0
package/cpp/src/unicode.cpp +1103 -0
package/cpp/src/unicode.h +111 -0
package/cpp/vendor/nlohmann/json.hpp +25526 -0
package/cpp/vendor/nlohmann/json_fwd.hpp +187 -0
package/cpp/vendor/stb/stb_image.h +7988 -0
package/ios/LocalLLM-Bridging-Header.h +2 -0
package/ios/LocalLLM.h +5 -0
package/ios/LocalLLM.mm +1267 -0
package/local-llm-rn.podspec +60 -0
package/package.json +35 -0
package/src/NativeLocalLLM.ts +73 -0
package/src/device.ts +50 -0
package/src/download-adapter.ts +17 -0
package/src/index.ts +21 -0
package/src/native-bridge.ts +142 -0
package/src/rn-downloader.ts +37 -0

package/cpp/src/llama-arch.cpp ADDED Viewed

@@ -0,0 +1,2855 @@
+#include "llama-arch.h"
+#include "llama-impl.h"
+#include <map>
+#include <set>
+static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
+    { LLM_ARCH_CLIP,             "clip"             }, // dummy, only used by llama-quantize
+    { LLM_ARCH_LLAMA,            "llama"            },
+    { LLM_ARCH_LLAMA4,           "llama4"           },
+    { LLM_ARCH_DECI,             "deci"             },
+    { LLM_ARCH_FALCON,           "falcon"           },
+    { LLM_ARCH_GROK,             "grok"             },
+    { LLM_ARCH_GPT2,             "gpt2"             },
+    { LLM_ARCH_GPTJ,             "gptj"             },
+    { LLM_ARCH_GPTNEOX,          "gptneox"          },
+    { LLM_ARCH_MPT,              "mpt"              },
+    { LLM_ARCH_BAICHUAN,         "baichuan"         },
+    { LLM_ARCH_STARCODER,        "starcoder"        },
+    { LLM_ARCH_REFACT,           "refact"           },
+    { LLM_ARCH_BERT,             "bert"             },
+    { LLM_ARCH_MODERN_BERT,      "modern-bert"      },
+    { LLM_ARCH_NOMIC_BERT,       "nomic-bert"       },
+    { LLM_ARCH_NOMIC_BERT_MOE,   "nomic-bert-moe"   },
+    { LLM_ARCH_NEO_BERT,         "neo-bert"         },
+    { LLM_ARCH_JINA_BERT_V2,     "jina-bert-v2"     },
+    { LLM_ARCH_JINA_BERT_V3,     "jina-bert-v3"     },
+    { LLM_ARCH_EUROBERT,         "eurobert"         },
+    { LLM_ARCH_BLOOM,            "bloom"            },
+    { LLM_ARCH_STABLELM,         "stablelm"         },
+    { LLM_ARCH_QWEN,             "qwen"             },
+    { LLM_ARCH_QWEN2,            "qwen2"            },
+    { LLM_ARCH_QWEN2MOE,         "qwen2moe"         },
+    { LLM_ARCH_QWEN2VL,          "qwen2vl"          },
+    { LLM_ARCH_QWEN3,            "qwen3"            },
+    { LLM_ARCH_QWEN3MOE,         "qwen3moe"         },
+    { LLM_ARCH_QWEN3NEXT,        "qwen3next"        },
+    { LLM_ARCH_QWEN3VL,          "qwen3vl"          },
+    { LLM_ARCH_QWEN3VLMOE,       "qwen3vlmoe"       },
+    { LLM_ARCH_QWEN35,           "qwen35"           },
+    { LLM_ARCH_QWEN35MOE,        "qwen35moe"        },
+    { LLM_ARCH_PHI2,             "phi2"             },
+    { LLM_ARCH_PHI3,             "phi3"             },
+    { LLM_ARCH_PHIMOE,           "phimoe"           },
+    { LLM_ARCH_PLAMO,            "plamo"            },
+    { LLM_ARCH_PLAMO2,           "plamo2"           },
+    { LLM_ARCH_PLAMO3,           "plamo3"           },
+    { LLM_ARCH_CODESHELL,        "codeshell"        },
+    { LLM_ARCH_ORION,            "orion"            },
+    { LLM_ARCH_INTERNLM2,        "internlm2"        },
+    { LLM_ARCH_MINICPM,          "minicpm"          },
+    { LLM_ARCH_MINICPM3,         "minicpm3"         },
+    { LLM_ARCH_GEMMA,            "gemma"            },
+    { LLM_ARCH_GEMMA2,           "gemma2"           },
+    { LLM_ARCH_GEMMA3,           "gemma3"           },
+    { LLM_ARCH_GEMMA3N,          "gemma3n"          },
+    { LLM_ARCH_GEMMA_EMBEDDING,  "gemma-embedding"  },
+    { LLM_ARCH_STARCODER2,       "starcoder2"       },
+    { LLM_ARCH_MAMBA,            "mamba"            },
+    { LLM_ARCH_MAMBA2,           "mamba2"           },
+    { LLM_ARCH_JAMBA,            "jamba"            },
+    { LLM_ARCH_FALCON_H1,        "falcon-h1"        },
+    { LLM_ARCH_XVERSE,           "xverse"           },
+    { LLM_ARCH_COMMAND_R,        "command-r"        },
+    { LLM_ARCH_COHERE2,          "cohere2"          },
+    { LLM_ARCH_DBRX,             "dbrx"             },
+    { LLM_ARCH_OLMO,             "olmo"             },
+    { LLM_ARCH_OLMO2,            "olmo2"            },
+    { LLM_ARCH_OLMOE,            "olmoe"            },
+    { LLM_ARCH_OPENELM,          "openelm"          },
+    { LLM_ARCH_ARCTIC,           "arctic"           },
+    { LLM_ARCH_DEEPSEEK,         "deepseek"         },
+    { LLM_ARCH_DEEPSEEK2,        "deepseek2"        },
+    { LLM_ARCH_CHATGLM,          "chatglm"          },
+    { LLM_ARCH_GLM4,             "glm4"             },
+    { LLM_ARCH_GLM4_MOE,         "glm4moe"          },
+    { LLM_ARCH_GLM_DSA,          "glm-dsa"          },
+    { LLM_ARCH_BITNET,           "bitnet"           },
+    { LLM_ARCH_T5,               "t5"               },
+    { LLM_ARCH_T5ENCODER,        "t5encoder"        },
+    { LLM_ARCH_JAIS,             "jais"             },
+    { LLM_ARCH_JAIS2,            "jais2"            },
+    { LLM_ARCH_NEMOTRON,         "nemotron"         },
+    { LLM_ARCH_NEMOTRON_H,       "nemotron_h"       },
+    { LLM_ARCH_NEMOTRON_H_MOE,   "nemotron_h_moe"   },
+    { LLM_ARCH_EXAONE,           "exaone"           },
+    { LLM_ARCH_EXAONE4,          "exaone4"          },
+    { LLM_ARCH_EXAONE_MOE,       "exaone-moe"       },
+    { LLM_ARCH_RWKV6,            "rwkv6"            },
+    { LLM_ARCH_RWKV6QWEN2,       "rwkv6qwen2"       },
+    { LLM_ARCH_RWKV7,            "rwkv7"            },
+    { LLM_ARCH_ARWKV7,           "arwkv7"           },
+    { LLM_ARCH_GRANITE,          "granite"          },
+    { LLM_ARCH_GRANITE_MOE,      "granitemoe"       },
+    { LLM_ARCH_GRANITE_HYBRID,   "granitehybrid"    },
+    { LLM_ARCH_CHAMELEON,        "chameleon"        },
+    { LLM_ARCH_WAVTOKENIZER_DEC, "wavtokenizer-dec" },
+    { LLM_ARCH_PLM,              "plm"              },
+    { LLM_ARCH_BAILINGMOE,       "bailingmoe"       },
+    { LLM_ARCH_BAILINGMOE2,      "bailingmoe2"      },
+    { LLM_ARCH_DOTS1,            "dots1"            },
+    { LLM_ARCH_ARCEE,            "arcee"            },
+    { LLM_ARCH_AFMOE,            "afmoe"            },
+    { LLM_ARCH_ERNIE4_5,         "ernie4_5"         },
+    { LLM_ARCH_ERNIE4_5_MOE,     "ernie4_5-moe"     },
+    { LLM_ARCH_HUNYUAN_MOE,      "hunyuan-moe"      },
+    { LLM_ARCH_HUNYUAN_DENSE,    "hunyuan-dense"    },
+    { LLM_ARCH_SMOLLM3,          "smollm3"          },
+    { LLM_ARCH_OPENAI_MOE,       "gpt-oss"          },
+    { LLM_ARCH_LFM2,             "lfm2"             },
+    { LLM_ARCH_LFM2MOE,          "lfm2moe"          },
+    { LLM_ARCH_DREAM,            "dream"            },
+    { LLM_ARCH_SMALLTHINKER,     "smallthinker"     },
+    { LLM_ARCH_LLADA,            "llada"            },
+    { LLM_ARCH_LLADA_MOE,        "llada-moe"        },
+    { LLM_ARCH_SEED_OSS,         "seed_oss"         },
+    { LLM_ARCH_GROVEMOE,         "grovemoe"         },
+    { LLM_ARCH_APERTUS,          "apertus"          },
+    { LLM_ARCH_MINIMAX_M2,       "minimax-m2"       },
+    { LLM_ARCH_COGVLM,           "cogvlm"           },
+    { LLM_ARCH_RND1,             "rnd1"             },
+    { LLM_ARCH_PANGU_EMBED,      "pangu-embedded"   },
+    { LLM_ARCH_MISTRAL3,         "mistral3"         },
+    { LLM_ARCH_PADDLEOCR,        "paddleocr"        },
+    { LLM_ARCH_MIMO2,            "mimo2"            },
+    { LLM_ARCH_STEP35,           "step35"           },
+    { LLM_ARCH_LLAMA_EMBED,      "llama-embed"      },
+    { LLM_ARCH_MAINCODER,        "maincoder"        },
+    { LLM_ARCH_KIMI_LINEAR,      "kimi-linear"      },
+    { LLM_ARCH_UNKNOWN,          "(unknown)"        },
+};
+static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
+    { LLM_KV_GENERAL_TYPE,                     "general.type"                          },
+    { LLM_KV_GENERAL_ARCHITECTURE,             "general.architecture"                  },
+    { LLM_KV_GENERAL_QUANTIZATION_VERSION,     "general.quantization_version"          },
+    { LLM_KV_GENERAL_ALIGNMENT,                "general.alignment"                     },
+    { LLM_KV_GENERAL_FILE_TYPE,                "general.file_type"                     },
+    { LLM_KV_GENERAL_SAMPLING_SEQUENCE,        "general.sampling.sequence"             },
+    { LLM_KV_GENERAL_SAMPLING_TOP_K,           "general.sampling.top_k"                },
+    { LLM_KV_GENERAL_SAMPLING_TOP_P,           "general.sampling.top_p"                },
+    { LLM_KV_GENERAL_SAMPLING_MIN_P,           "general.sampling.min_p"                },
+    { LLM_KV_GENERAL_SAMPLING_XTC_PROBABILITY, "general.sampling.xtc_probability"      },
+    { LLM_KV_GENERAL_SAMPLING_XTC_THRESHOLD,   "general.sampling.xtc_threshold"        },
+    { LLM_KV_GENERAL_SAMPLING_TEMP,            "general.sampling.temp"                 },
+    { LLM_KV_GENERAL_SAMPLING_PENALTY_LAST_N,  "general.sampling.penalty_last_n"       },
+    { LLM_KV_GENERAL_SAMPLING_PENALTY_REPEAT,  "general.sampling.penalty_repeat"       },
+    { LLM_KV_GENERAL_SAMPLING_MIROSTAT,        "general.sampling.mirostat"             },
+    { LLM_KV_GENERAL_SAMPLING_MIROSTAT_TAU,    "general.sampling.mirostat_tau"         },
+    { LLM_KV_GENERAL_SAMPLING_MIROSTAT_ETA,    "general.sampling.mirostat_eta"         },
+    { LLM_KV_GENERAL_NAME,                     "general.name"                          },
+    { LLM_KV_GENERAL_AUTHOR,                   "general.author"                        },
+    { LLM_KV_GENERAL_VERSION,                  "general.version"                       },
+    { LLM_KV_GENERAL_URL,                      "general.url"                           },
+    { LLM_KV_GENERAL_DESCRIPTION,              "general.description"                   },
+    { LLM_KV_GENERAL_LICENSE,                  "general.license"                       },
+    { LLM_KV_GENERAL_SOURCE_URL,               "general.source.url"                    },
+    { LLM_KV_GENERAL_SOURCE_HF_REPO,           "general.source.huggingface.repository" },
+    { LLM_KV_VOCAB_SIZE,                        "%s.vocab_size"                        },
+    { LLM_KV_CONTEXT_LENGTH,                    "%s.context_length"                    },
+    { LLM_KV_EMBEDDING_LENGTH,                  "%s.embedding_length"                  },
+    { LLM_KV_EMBEDDING_LENGTH_OUT,              "%s.embedding_length_out"              },
+    { LLM_KV_FEATURES_LENGTH,                   "%s.features_length"                   },
+    { LLM_KV_BLOCK_COUNT,                       "%s.block_count"                       },
+    { LLM_KV_LEADING_DENSE_BLOCK_COUNT,         "%s.leading_dense_block_count"         },
+    { LLM_KV_FEED_FORWARD_LENGTH,               "%s.feed_forward_length"               },
+    { LLM_KV_EXPERT_FEED_FORWARD_LENGTH,        "%s.expert_feed_forward_length"        },
+    { LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, "%s.expert_shared_feed_forward_length" },
+    { LLM_KV_EXPERT_CHUNK_FEED_FORWARD_LENGTH,  "%s.expert_chunk_feed_forward_length"  },
+    { LLM_KV_SWIGLU_CLAMP_EXP,                  "%s.swiglu_clamp_exp"                  },
+    { LLM_KV_SWIGLU_CLAMP_SHEXP,                "%s.swiglu_clamp_shexp"                },
+    { LLM_KV_USE_PARALLEL_RESIDUAL,             "%s.use_parallel_residual"             },
+    { LLM_KV_TENSOR_DATA_LAYOUT,                "%s.tensor_data_layout"                },
+    { LLM_KV_EXPERT_COUNT,                      "%s.expert_count"                      },
+    { LLM_KV_EXPERT_USED_COUNT,                 "%s.expert_used_count"                 },
+    { LLM_KV_EXPERT_SHARED_COUNT,               "%s.expert_shared_count"               },
+    { LLM_KV_EXPERT_GROUP_COUNT,                "%s.expert_group_count"                },
+    { LLM_KV_EXPERT_GROUP_USED_COUNT,           "%s.expert_group_used_count"           },
+    { LLM_KV_EXPERT_WEIGHTS_SCALE,              "%s.expert_weights_scale"              },
+    { LLM_KV_EXPERT_WEIGHTS_NORM,               "%s.expert_weights_norm"               },
+    { LLM_KV_EXPERT_GATING_FUNC,                "%s.expert_gating_func"                },
+    { LLM_KV_EXPERT_GROUP_SCALE,                "%s.expert_group_scale"                },
+    { LLM_KV_EXPERTS_PER_GROUP,                 "%s.experts_per_group"                 },
+    { LLM_KV_MOE_EVERY_N_LAYERS,                "%s.moe_every_n_layers"                },
+    { LLM_KV_NEXTN_PREDICT_LAYERS,              "%s.nextn_predict_layers"              },
+    { LLM_KV_NUM_DEEPSTACK_LAYERS,              "%s.n_deepstack_layers"                },
+    { LLM_KV_POOLING_TYPE,                      "%s.pooling_type"                      },
+    { LLM_KV_LOGIT_SCALE,                       "%s.logit_scale"                       },
+    { LLM_KV_DECODER_START_TOKEN_ID,            "%s.decoder_start_token_id"            },
+    { LLM_KV_DECODER_BLOCK_COUNT,               "%s.decoder_block_count"               },
+    { LLM_KV_ATTN_LOGIT_SOFTCAPPING,            "%s.attn_logit_softcapping"            },
+    { LLM_KV_ROUTER_LOGIT_SOFTCAPPING,          "%s.router_logit_softcapping"          },
+    { LLM_KV_FINAL_LOGIT_SOFTCAPPING,           "%s.final_logit_softcapping"           },
+    { LLM_KV_SWIN_NORM,                         "%s.swin_norm"                         },
+    { LLM_KV_RESCALE_EVERY_N_LAYERS,            "%s.rescale_every_n_layers"            },
+    { LLM_KV_TIME_MIX_EXTRA_DIM,                "%s.time_mix_extra_dim"                },
+    { LLM_KV_TIME_DECAY_EXTRA_DIM,              "%s.time_decay_extra_dim"              },
+    { LLM_KV_RESIDUAL_SCALE,                    "%s.residual_scale"                    },
+    { LLM_KV_EMBEDDING_SCALE,                   "%s.embedding_scale"                   },
+    { LLM_KV_TOKEN_SHIFT_COUNT,                 "%s.token_shift_count"                 },
+    { LLM_KV_INTERLEAVE_MOE_LAYER_STEP,         "%s.interleave_moe_layer_step"         },
+    { LLM_KV_FULL_ATTENTION_INTERVAL,           "%s.full_attention_interval"           },
+    { LLM_KV_ATTENTION_HEAD_COUNT,                   "%s.attention.head_count"                   },
+    { LLM_KV_ATTENTION_HEAD_COUNT_KV,                "%s.attention.head_count_kv"                },
+    { LLM_KV_ATTENTION_MAX_ALIBI_BIAS,               "%s.attention.max_alibi_bias"               },
+    { LLM_KV_ATTENTION_CLAMP_KQV,                    "%s.attention.clamp_kqv"                    },
+    { LLM_KV_ATTENTION_KEY_LENGTH,                   "%s.attention.key_length"                   },
+    { LLM_KV_ATTENTION_VALUE_LENGTH,                 "%s.attention.value_length"                 },
+    { LLM_KV_ATTENTION_LAYERNORM_EPS,                "%s.attention.layer_norm_epsilon"           },
+    { LLM_KV_ATTENTION_LAYERNORM_RMS_EPS,            "%s.attention.layer_norm_rms_epsilon"       },
+    { LLM_KV_ATTENTION_GROUPNORM_EPS,                "%s.attention.group_norm_epsilon"           },
+    { LLM_KV_ATTENTION_GROUPNORM_GROUPS,             "%s.attention.group_norm_groups"            },
+    { LLM_KV_ATTENTION_CAUSAL,                       "%s.attention.causal"                       },
+    { LLM_KV_ATTENTION_Q_LORA_RANK,                  "%s.attention.q_lora_rank"                  },
+    { LLM_KV_ATTENTION_KV_LORA_RANK,                 "%s.attention.kv_lora_rank"                 },
+    { LLM_KV_ATTENTION_DECAY_LORA_RANK,              "%s.attention.decay_lora_rank"              },
+    { LLM_KV_ATTENTION_ICLR_LORA_RANK,               "%s.attention.iclr_lora_rank"               },
+    { LLM_KV_ATTENTION_VALUE_RESIDUAL_MIX_LORA_RANK, "%s.attention.value_residual_mix_lora_rank" },
+    { LLM_KV_ATTENTION_GATE_LORA_RANK,               "%s.attention.gate_lora_rank"               },
+    { LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT,       "%s.attention.relative_buckets_count"       },
+    { LLM_KV_ATTENTION_SLIDING_WINDOW,               "%s.attention.sliding_window"               },
+    { LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN,       "%s.attention.sliding_window_pattern"       },
+    { LLM_KV_ATTENTION_SCALE,                        "%s.attention.scale"                        },
+    { LLM_KV_ATTENTION_OUTPUT_SCALE,                 "%s.attention.output_scale"                 },
+    { LLM_KV_ATTENTION_TEMPERATURE_LENGTH,           "%s.attention.temperature_length"           },
+    { LLM_KV_ATTENTION_TEMPERATURE_SCALE,            "%s.attention.temperature_scale"            },
+    { LLM_KV_ATTENTION_KEY_LENGTH_MLA,               "%s.attention.key_length_mla"               },
+    { LLM_KV_ATTENTION_VALUE_LENGTH_MLA,             "%s.attention.value_length_mla"             },
+    { LLM_KV_ATTENTION_INDEXER_HEAD_COUNT,           "%s.attention.indexer.head_count"           },
+    { LLM_KV_ATTENTION_INDEXER_KEY_LENGTH,           "%s.attention.indexer.key_length"           },
+    { LLM_KV_ATTENTION_INDEXER_TOP_K,                "%s.attention.indexer.top_k"                },
+    { LLM_KV_ROPE_DIMENSION_COUNT,           "%s.rope.dimension_count"                 },
+    { LLM_KV_ROPE_DIMENSION_SECTIONS,        "%s.rope.dimension_sections"              },
+    { LLM_KV_ROPE_FREQ_BASE,                 "%s.rope.freq_base"                       },
+    { LLM_KV_ROPE_FREQ_BASE_SWA,             "%s.rope.freq_base_swa"                   },
+    { LLM_KV_ROPE_SCALE_LINEAR,              "%s.rope.scale_linear"                    },
+    { LLM_KV_ROPE_SCALING_TYPE,              "%s.rope.scaling.type"                    },
+    { LLM_KV_ROPE_SCALING_FACTOR,            "%s.rope.scaling.factor"                  },
+    { LLM_KV_ROPE_SCALING_ATTN_FACTOR,       "%s.rope.scaling.attn_factor"             },
+    { LLM_KV_ROPE_SCALING_ORIG_CTX_LEN,      "%s.rope.scaling.original_context_length" },
+    { LLM_KV_ROPE_SCALING_FINETUNED,         "%s.rope.scaling.finetuned"               },
+    { LLM_KV_ROPE_SCALING_YARN_LOG_MUL,      "%s.rope.scaling.yarn_log_multiplier"     },
+    { LLM_KV_ROPE_SCALING_YARN_EXT_FACTOR,   "%s.rope.scaling.yarn_ext_factor"         },
+    { LLM_KV_ROPE_SCALING_YARN_ATTN_FACTOR,  "%s.rope.scaling.yarn_attn_factor"        },
+    { LLM_KV_ROPE_SCALING_YARN_BETA_FAST,    "%s.rope.scaling.yarn_beta_fast"          },
+    { LLM_KV_ROPE_SCALING_YARN_BETA_SLOW,    "%s.rope.scaling.yarn_beta_slow"          },
+    { LLM_KV_SPLIT_NO,            "split.no"            },
+    { LLM_KV_SPLIT_COUNT,         "split.count"         },
+    { LLM_KV_SPLIT_TENSORS_COUNT, "split.tensors.count" },
+    { LLM_KV_SSM_CONV_KERNEL,    "%s.ssm.conv_kernel"    },
+    { LLM_KV_SSM_INNER_SIZE,     "%s.ssm.inner_size"     },
+    { LLM_KV_SSM_STATE_SIZE,     "%s.ssm.state_size"     },
+    { LLM_KV_SSM_TIME_STEP_RANK, "%s.ssm.time_step_rank" },
+    { LLM_KV_SSM_GROUP_COUNT,    "%s.ssm.group_count"    },
+    { LLM_KV_SSM_DT_B_C_RMS,     "%s.ssm.dt_b_c_rms"     },
+    { LLM_KV_KDA_HEAD_DIM, "%s.kda.head_dim" },
+    { LLM_KV_WKV_HEAD_SIZE, "%s.wkv.head_size" },
+    { LLM_KV_POSNET_EMBEDDING_LENGTH, "%s.posnet.embedding_length" },
+    { LLM_KV_POSNET_BLOCK_COUNT,      "%s.posnet.block_count"      },
+    { LLM_KV_CONVNEXT_EMBEDDING_LENGTH, "%s.convnext.embedding_length" },
+    { LLM_KV_CONVNEXT_BLOCK_COUNT,      "%s.convnext.block_count"      },
+    { LLM_KV_CLASSIFIER_OUTPUT_LABELS, "%s.classifier.output_labels" },
+    { LLM_KV_SHORTCONV_L_CACHE, "%s.shortconv.l_cache" },
+    // sentence-transformers dense modules feature dims
+    { LLM_KV_DENSE_2_FEAT_IN,        "%s.dense_2_feat_in"  },
+    { LLM_KV_DENSE_2_FEAT_OUT,       "%s.dense_2_feat_out"  },
+    { LLM_KV_DENSE_3_FEAT_IN,        "%s.dense_3_feat_in"   },
+    { LLM_KV_DENSE_3_FEAT_OUT,       "%s.dense_3_feat_out"  },
+    { LLM_KV_TOKENIZER_MODEL,                "tokenizer.ggml.model"                    },
+    { LLM_KV_TOKENIZER_PRE,                  "tokenizer.ggml.pre"                      },
+    { LLM_KV_TOKENIZER_LIST,                 "tokenizer.ggml.tokens"                   },
+    { LLM_KV_TOKENIZER_TOKEN_TYPE,           "tokenizer.ggml.token_type"               },
+    { LLM_KV_TOKENIZER_TOKEN_TYPE_COUNT,     "tokenizer.ggml.token_type_count"         },
+    { LLM_KV_TOKENIZER_SCORES,               "tokenizer.ggml.scores"                   },
+    { LLM_KV_TOKENIZER_MERGES,               "tokenizer.ggml.merges"                   },
+    { LLM_KV_TOKENIZER_BOS_ID,               "tokenizer.ggml.bos_token_id"             },
+    { LLM_KV_TOKENIZER_EOS_ID,               "tokenizer.ggml.eos_token_id"             },
+    { LLM_KV_TOKENIZER_EOT_ID,               "tokenizer.ggml.eot_token_id"             },
+    { LLM_KV_TOKENIZER_EOM_ID,               "tokenizer.ggml.eom_token_id"             },
+    { LLM_KV_TOKENIZER_UNK_ID,               "tokenizer.ggml.unknown_token_id"         },
+    { LLM_KV_TOKENIZER_SEP_ID,               "tokenizer.ggml.seperator_token_id"       },
+    { LLM_KV_TOKENIZER_PAD_ID,               "tokenizer.ggml.padding_token_id"         },
+    { LLM_KV_TOKENIZER_CLS_ID,               "tokenizer.ggml.cls_token_id"             },
+    { LLM_KV_TOKENIZER_MASK_ID,              "tokenizer.ggml.mask_token_id"            },
+    { LLM_KV_TOKENIZER_ADD_BOS,              "tokenizer.ggml.add_bos_token"            },
+    { LLM_KV_TOKENIZER_ADD_EOS,              "tokenizer.ggml.add_eos_token"            },
+    { LLM_KV_TOKENIZER_ADD_SEP,              "tokenizer.ggml.add_sep_token"            },
+    { LLM_KV_TOKENIZER_ADD_PREFIX,           "tokenizer.ggml.add_space_prefix"         },
+    { LLM_KV_TOKENIZER_REMOVE_EXTRA_WS,      "tokenizer.ggml.remove_extra_whitespaces" },
+    { LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, "tokenizer.ggml.precompiled_charsmap"     },
+    { LLM_KV_TOKENIZER_HF_JSON,              "tokenizer.huggingface.json"              },
+    { LLM_KV_TOKENIZER_RWKV,                 "tokenizer.rwkv.world"                    },
+    { LLM_KV_TOKENIZER_CHAT_TEMPLATE,        "tokenizer.chat_template"                 },
+    { LLM_KV_TOKENIZER_FIM_PRE_ID,           "tokenizer.ggml.fim_pre_token_id"         },
+    { LLM_KV_TOKENIZER_FIM_SUF_ID,           "tokenizer.ggml.fim_suf_token_id"         },
+    { LLM_KV_TOKENIZER_FIM_MID_ID,           "tokenizer.ggml.fim_mid_token_id"         },
+    { LLM_KV_TOKENIZER_FIM_PAD_ID,           "tokenizer.ggml.fim_pad_token_id"         },
+    { LLM_KV_TOKENIZER_FIM_REP_ID,           "tokenizer.ggml.fim_rep_token_id"         },
+    { LLM_KV_TOKENIZER_FIM_SEP_ID,           "tokenizer.ggml.fim_sep_token_id"         },
+    { LLM_KV_ADAPTER_TYPE,                    "adapter.type"               },
+    { LLM_KV_ADAPTER_LORA_ALPHA,              "adapter.lora.alpha"         },
+    { LLM_KV_ADAPTER_LORA_TASK_NAME,          "adapter.lora.task_name"     },
+    { LLM_KV_ADAPTER_LORA_PROMPT_PREFIX,      "adapter.lora.prompt_prefix" },
+    { LLM_KV_ADAPTER_ALORA_INVOCATION_TOKENS, "adapter.alora.invocation_tokens" },
+    { LLM_KV_XIELU_ALPHA_N,         "xielu.alpha_n"         },
+    { LLM_KV_XIELU_ALPHA_P,         "xielu.alpha_p"         },
+    { LLM_KV_XIELU_BETA,            "xielu.beta"            },
+    { LLM_KV_XIELU_EPS,             "xielu.eps"             },
+    // deprecated
+    { LLM_KV_TOKENIZER_PREFIX_ID, "tokenizer.ggml.prefix_token_id" },
+    { LLM_KV_TOKENIZER_SUFFIX_ID, "tokenizer.ggml.suffix_token_id" },
+    { LLM_KV_TOKENIZER_MIDDLE_ID, "tokenizer.ggml.middle_token_id" },
+};
+static const std::map<llm_tensor, const char *> LLM_TENSOR_NAMES = {
+    { LLM_TENSOR_TOKEN_EMBD,                             "token_embd" },
+    { LLM_TENSOR_OUTPUT_NORM,                            "output_norm" },
+    { LLM_TENSOR_OUTPUT_NORM_LFM2,                       "token_embd_norm" }, // fix for wrong tensor name
+    { LLM_TENSOR_OUTPUT,                                 "output" },
+    { LLM_TENSOR_ROPE_FREQS,                             "rope_freqs" },
+    { LLM_TENSOR_ATTN_NORM,                              "blk.%d.attn_norm" },
+    { LLM_TENSOR_ATTN_Q,                                 "blk.%d.attn_q" },
+    { LLM_TENSOR_ATTN_K,                                 "blk.%d.attn_k" },
+    { LLM_TENSOR_ATTN_V,                                 "blk.%d.attn_v" },
+    { LLM_TENSOR_ATTN_OUT,                               "blk.%d.attn_output" },
+    { LLM_TENSOR_ATTN_ROT_EMBD,                          "blk.%d.attn_rot_embd" },
+    { LLM_TENSOR_FFN_GATE_INP,                           "blk.%d.ffn_gate_inp" },
+    { LLM_TENSOR_FFN_NORM,                               "blk.%d.ffn_norm" },
+    { LLM_TENSOR_FFN_GATE,                               "blk.%d.ffn_gate" },
+    { LLM_TENSOR_FFN_DOWN,                               "blk.%d.ffn_down" },
+    { LLM_TENSOR_FFN_UP,                                 "blk.%d.ffn_up" },
+    { LLM_TENSOR_FFN_GATE_EXP,                           "blk.%d.ffn_gate.%d" },
+    { LLM_TENSOR_FFN_DOWN_EXP,                           "blk.%d.ffn_down.%d" },
+    { LLM_TENSOR_FFN_UP_EXP,                             "blk.%d.ffn_up.%d" },
+    { LLM_TENSOR_FFN_GATE_EXPS,                          "blk.%d.ffn_gate_exps" },
+    { LLM_TENSOR_FFN_GATE_UP_EXPS,                       "blk.%d.ffn_gate_up_exps" },
+    { LLM_TENSOR_FFN_DOWN_EXPS,                          "blk.%d.ffn_down_exps" },
+    { LLM_TENSOR_FFN_UP_EXPS,                            "blk.%d.ffn_up_exps" },
+    { LLM_TENSOR_ATTN_POST_NORM,                         "blk.%d.post_attention_norm" },
+    { LLM_TENSOR_ATTN_Q_NORM,                            "blk.%d.attn_q_norm" },
+    { LLM_TENSOR_ATTN_K_NORM,                            "blk.%d.attn_k_norm" },
+    { LLM_TENSOR_ATTN_GATE,                              "blk.%d.attn_gate" },
+    { LLM_TENSOR_FFN_POST_NORM,                          "blk.%d.post_ffw_norm" },
+    { LLM_TENSOR_FFN_GATE_SHEXP,                         "blk.%d.ffn_gate_shexp" },
+    { LLM_TENSOR_FFN_UP_SHEXP,                           "blk.%d.ffn_up_shexp" },
+    { LLM_TENSOR_FFN_DOWN_SHEXP,                         "blk.%d.ffn_down_shexp" },
+    { LLM_TENSOR_FFN_EXP_PROBS_B,                        "blk.%d.exp_probs_b" },
+    { LLM_TENSOR_ATTN_NORM_2,                            "blk.%d.attn_norm_2" },
+    { LLM_TENSOR_ATTN_QKV,                               "blk.%d.attn_qkv" },
+    { LLM_TENSOR_LAYER_OUT_NORM,                         "blk.%d.layer_output_norm" },
+    { LLM_TENSOR_ATTN_OUT_NORM,                          "blk.%d.attn_output_norm" },
+    { LLM_TENSOR_POS_EMBD,                               "position_embd" },
+    { LLM_TENSOR_FFN_ACT,                                "blk.%d.ffn.act" },
+    { LLM_TENSOR_TOKEN_EMBD_NORM,                        "token_embd_norm" },
+    { LLM_TENSOR_TOKEN_TYPES,                            "token_types" },
+    { LLM_TENSOR_CLS,                                    "cls" },
+    { LLM_TENSOR_CLS_OUT,                                "cls.output" },
+    { LLM_TENSOR_CLS_NORM,                               "cls.norm" },
+    { LLM_TENSOR_ENC_OUTPUT_NORM,                        "enc.output_norm" },
+    { LLM_TENSOR_FFN_GATE_INP_SHEXP,                     "blk.%d.ffn_gate_inp_shexp" },
+    { LLM_TENSOR_SSM_A_NOSCAN,                           "blk.%d.ssm_a" },
+    { LLM_TENSOR_SSM_CONV1D,                             "blk.%d.ssm_conv1d" },
+    { LLM_TENSOR_SSM_DT,                                 "blk.%d.ssm_dt" },
+    { LLM_TENSOR_SSM_BETA_ALPHA,                         "blk.%d.ssm_ba" },
+    { LLM_TENSOR_SSM_ALPHA,                              "blk.%d.ssm_alpha" },
+    { LLM_TENSOR_SSM_IN,                                 "blk.%d.ssm_in" },
+    { LLM_TENSOR_SSM_NORM,                               "blk.%d.ssm_norm" },
+    { LLM_TENSOR_SSM_OUT,                                "blk.%d.ssm_out" },
+    { LLM_TENSOR_ROPE_FACTORS_LONG,                      "rope_factors_long" },
+    { LLM_TENSOR_ROPE_FACTORS_SHORT,                     "rope_factors_short" },
+    { LLM_TENSOR_SSM_X,                                  "blk.%d.ssm_x" },
+    { LLM_TENSOR_SSM_A,                                  "blk.%d.ssm_a" },
+    { LLM_TENSOR_SSM_D,                                  "blk.%d.ssm_d" },
+    { LLM_TENSOR_SSM_DT_NORM,                            "blk.%d.ssm_dt_norm" },
+    { LLM_TENSOR_SSM_B_NORM,                             "blk.%d.ssm_b_norm" },
+    { LLM_TENSOR_SSM_C_NORM,                             "blk.%d.ssm_c_norm" },
+    { LLM_TENSOR_SSM_CONV1D_Q,                           "blk.%d.ssm_conv1d_q" },
+    { LLM_TENSOR_SSM_CONV1D_K,                           "blk.%d.ssm_conv1d_k" },
+    { LLM_TENSOR_SSM_CONV1D_V,                           "blk.%d.ssm_conv1d_v" },
+    { LLM_TENSOR_SSM_F_A,                                "blk.%d.ssm_f_a" },
+    { LLM_TENSOR_SSM_F_B,                                "blk.%d.ssm_f_b" },
+    { LLM_TENSOR_SSM_BETA,                               "blk.%d.ssm_beta" },
+    { LLM_TENSOR_SSM_G_A,                                "blk.%d.ssm_g_a" },
+    { LLM_TENSOR_SSM_G_B,                                "blk.%d.ssm_g_b" },
+    { LLM_TENSOR_SSM_NORM,                               "blk.%d.ssm_norm" },
+    { LLM_TENSOR_ATTN_Q_A_NORM,                          "blk.%d.attn_q_a_norm" },
+    { LLM_TENSOR_ATTN_KV_A_NORM,                         "blk.%d.attn_kv_a_norm" },
+    { LLM_TENSOR_ATTN_Q_A,                               "blk.%d.attn_q_a" },
+    { LLM_TENSOR_ATTN_Q_B,                               "blk.%d.attn_q_b" },
+    { LLM_TENSOR_ATTN_KV_A_MQA,                          "blk.%d.attn_kv_a_mqa" },
+    { LLM_TENSOR_ATTN_KV_B,                              "blk.%d.attn_kv_b" },
+    { LLM_TENSOR_PER_LAYER_TOKEN_EMBD,                   "per_layer_token_embd" },
+    { LLM_TENSOR_PER_LAYER_MODEL_PROJ,                   "per_layer_model_proj" },
+    { LLM_TENSOR_PER_LAYER_PROJ_NORM,                    "per_layer_proj_norm" },
+    { LLM_TENSOR_ALTUP_UNEMBD_PROJ,                      "altup_unembd_proj" },
+    { LLM_TENSOR_ALTUP_PROJ,                             "altup_proj" },
+    { LLM_TENSOR_PER_LAYER_INP_GATE,                     "blk.%d.inp_gate" },
+    { LLM_TENSOR_PER_LAYER_PROJ,                         "blk.%d.proj" },
+    { LLM_TENSOR_PER_LAYER_POST_NORM,                    "blk.%d.post_norm" },
+    { LLM_TENSOR_ALTUP_CORRECT_COEF,                     "blk.%d.altup_correct_coef" },
+    { LLM_TENSOR_ALTUP_CORRECT_SCALE,                    "blk.%d.altup_correct_scale" },
+    { LLM_TENSOR_ALTUP_PREDICT_COEF,                     "blk.%d.altup_predict_coef" },
+    { LLM_TENSOR_ALTUP_ROUTER,                           "blk.%d.altup_router" },
+    { LLM_TENSOR_ALTUP_ROUTER_NORM,                      "blk.%d.altup_router_norm" },
+    { LLM_TENSOR_LAUREL_L,                               "blk.%d.laurel_l" },
+    { LLM_TENSOR_LAUREL_R,                               "blk.%d.laurel_r" },
+    { LLM_TENSOR_LAUREL_POST_NORM,                       "blk.%d.laurel_post_norm" },
+    { LLM_TENSOR_DENSE_2_OUT,                            "dense_2" },
+    { LLM_TENSOR_DENSE_3_OUT,                            "dense_3" },
+    { LLM_TENSOR_FFN_NORM_EXPS,                          "blk.%d.ffn_norm_exps" },
+    { LLM_TENSOR_ATTN_K_B,                               "blk.%d.attn_k_b" },
+    { LLM_TENSOR_ATTN_V_B,                               "blk.%d.attn_v_b" },
+    { LLM_TENSOR_NEXTN_EH_PROJ,                          "blk.%d.nextn.eh_proj" },
+    { LLM_TENSOR_NEXTN_EMBED_TOKENS,                     "blk.%d.nextn.embed_tokens" },
+    { LLM_TENSOR_NEXTN_ENORM,                            "blk.%d.nextn.enorm" },
+    { LLM_TENSOR_NEXTN_HNORM,                            "blk.%d.nextn.hnorm" },
+    { LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,                 "blk.%d.nextn.shared_head_head" },
+    { LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,                 "blk.%d.nextn.shared_head_norm" },
+    { LLM_TENSOR_ATTN_SUB_NORM,                          "blk.%d.attn_sub_norm" },
+    { LLM_TENSOR_FFN_SUB_NORM,                           "blk.%d.ffn_sub_norm" },
+    { LLM_TENSOR_DEC_OUTPUT_NORM,                        "dec.output_norm" },
+    { LLM_TENSOR_DEC_ATTN_NORM,                          "dec.blk.%d.attn_norm" },
+    { LLM_TENSOR_DEC_ATTN_Q,                             "dec.blk.%d.attn_q" },
+    { LLM_TENSOR_DEC_ATTN_K,                             "dec.blk.%d.attn_k" },
+    { LLM_TENSOR_DEC_ATTN_V,                             "dec.blk.%d.attn_v" },
+    { LLM_TENSOR_DEC_ATTN_OUT,                           "dec.blk.%d.attn_o" },
+    { LLM_TENSOR_DEC_ATTN_REL_B,                         "dec.blk.%d.attn_rel_b" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_NORM,                    "dec.blk.%d.cross_attn_norm" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_Q,                       "dec.blk.%d.cross_attn_q" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_K,                       "dec.blk.%d.cross_attn_k" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_V,                       "dec.blk.%d.cross_attn_v" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_OUT,                     "dec.blk.%d.cross_attn_o" },
+    { LLM_TENSOR_DEC_CROSS_ATTN_REL_B,                   "dec.blk.%d.cross_attn_rel_b" },
+    { LLM_TENSOR_DEC_FFN_NORM,                           "dec.blk.%d.ffn_norm" },
+    { LLM_TENSOR_DEC_FFN_GATE,                           "dec.blk.%d.ffn_gate" },
+    { LLM_TENSOR_DEC_FFN_DOWN,                           "dec.blk.%d.ffn_down" },
+    { LLM_TENSOR_DEC_FFN_UP,                             "dec.blk.%d.ffn_up" },
+    { LLM_TENSOR_ENC_ATTN_NORM,                          "enc.blk.%d.attn_norm" },
+    { LLM_TENSOR_ENC_ATTN_Q,                             "enc.blk.%d.attn_q" },
+    { LLM_TENSOR_ENC_ATTN_K,                             "enc.blk.%d.attn_k" },
+    { LLM_TENSOR_ENC_ATTN_V,                             "enc.blk.%d.attn_v" },
+    { LLM_TENSOR_ENC_ATTN_OUT,                           "enc.blk.%d.attn_o" },
+    { LLM_TENSOR_ENC_ATTN_REL_B,                         "enc.blk.%d.attn_rel_b" },
+    { LLM_TENSOR_ENC_FFN_NORM,                           "enc.blk.%d.ffn_norm" },
+    { LLM_TENSOR_ENC_FFN_GATE,                           "enc.blk.%d.ffn_gate" },
+    { LLM_TENSOR_ENC_FFN_DOWN,                           "enc.blk.%d.ffn_down" },
+    { LLM_TENSOR_ENC_FFN_UP,                             "enc.blk.%d.ffn_up" },
+    { LLM_TENSOR_TIME_MIX_W1,                            "blk.%d.time_mix_w1" },
+    { LLM_TENSOR_TIME_MIX_W2,                            "blk.%d.time_mix_w2" },
+    { LLM_TENSOR_TIME_MIX_LERP_X,                        "blk.%d.time_mix_lerp_x" },
+    { LLM_TENSOR_TIME_MIX_LERP_W,                        "blk.%d.time_mix_lerp_w" },
+    { LLM_TENSOR_TIME_MIX_LERP_K,                        "blk.%d.time_mix_lerp_k" },
+    { LLM_TENSOR_TIME_MIX_LERP_V,                        "blk.%d.time_mix_lerp_v" },
+    { LLM_TENSOR_TIME_MIX_LERP_R,                        "blk.%d.time_mix_lerp_r" },
+    { LLM_TENSOR_TIME_MIX_LERP_G,                        "blk.%d.time_mix_lerp_g" },
+    { LLM_TENSOR_TIME_MIX_LERP_FUSED,                    "blk.%d.time_mix_lerp_fused" },
+    { LLM_TENSOR_TIME_MIX_FIRST,                         "blk.%d.time_mix_first" },
+    { LLM_TENSOR_TIME_MIX_DECAY,                         "blk.%d.time_mix_decay" },
+    { LLM_TENSOR_TIME_MIX_DECAY_W1,                      "blk.%d.time_mix_decay_w1" },
+    { LLM_TENSOR_TIME_MIX_DECAY_W2,                      "blk.%d.time_mix_decay_w2" },
+    { LLM_TENSOR_TIME_MIX_KEY,                           "blk.%d.time_mix_key" },
+    { LLM_TENSOR_TIME_MIX_VALUE,                         "blk.%d.time_mix_value" },
+    { LLM_TENSOR_TIME_MIX_RECEPTANCE,                    "blk.%d.time_mix_receptance" },
+    { LLM_TENSOR_TIME_MIX_GATE,                          "blk.%d.time_mix_gate" },
+    { LLM_TENSOR_TIME_MIX_LN,                            "blk.%d.time_mix_ln" },
+    { LLM_TENSOR_TIME_MIX_OUTPUT,                        "blk.%d.time_mix_output" },
+    { LLM_TENSOR_CHANNEL_MIX_LERP_K,                     "blk.%d.channel_mix_lerp_k" },
+    { LLM_TENSOR_CHANNEL_MIX_LERP_R,                     "blk.%d.channel_mix_lerp_r" },
+    { LLM_TENSOR_CHANNEL_MIX_KEY,                        "blk.%d.channel_mix_key" },
+    { LLM_TENSOR_CHANNEL_MIX_VALUE,                      "blk.%d.channel_mix_value" },
+    { LLM_TENSOR_CHANNEL_MIX_RECEPTANCE,                 "blk.%d.channel_mix_receptance" },
+    { LLM_TENSOR_TIME_MIX_W0,                            "blk.%d.time_mix_w0" },
+    { LLM_TENSOR_TIME_MIX_A0,                            "blk.%d.time_mix_a0" },
+    { LLM_TENSOR_TIME_MIX_A1,                            "blk.%d.time_mix_a1" },
+    { LLM_TENSOR_TIME_MIX_A2,                            "blk.%d.time_mix_a2" },
+    { LLM_TENSOR_TIME_MIX_V0,                            "blk.%d.time_mix_v0" },
+    { LLM_TENSOR_TIME_MIX_V1,                            "blk.%d.time_mix_v1" },
+    { LLM_TENSOR_TIME_MIX_V2,                            "blk.%d.time_mix_v2" },
+    { LLM_TENSOR_TIME_MIX_G1,                            "blk.%d.time_mix_g1" },
+    { LLM_TENSOR_TIME_MIX_G2,                            "blk.%d.time_mix_g2" },
+    { LLM_TENSOR_TIME_MIX_K_K,                           "blk.%d.time_mix_k_k" },
+    { LLM_TENSOR_TIME_MIX_K_A,                           "blk.%d.time_mix_k_a" },
+    { LLM_TENSOR_TIME_MIX_R_K,                           "blk.%d.time_mix_r_k" },
+    { LLM_TENSOR_CONV1D,                                 "conv1d" },
+    { LLM_TENSOR_CONVNEXT_DW,                            "convnext.%d.dw" },
+    { LLM_TENSOR_CONVNEXT_NORM,                          "convnext.%d.norm" },
+    { LLM_TENSOR_CONVNEXT_PW1,                           "convnext.%d.pw1" },
+    { LLM_TENSOR_CONVNEXT_PW2,                           "convnext.%d.pw2" },
+    { LLM_TENSOR_CONVNEXT_GAMMA,                         "convnext.%d.gamma" },
+    { LLM_TENSOR_POS_NET_CONV1,                          "posnet.%d.conv1" },
+    { LLM_TENSOR_POS_NET_CONV2,                          "posnet.%d.conv2" },
+    { LLM_TENSOR_POS_NET_NORM,                           "posnet.%d.norm" },
+    { LLM_TENSOR_POS_NET_NORM1,                          "posnet.%d.norm1" },
+    { LLM_TENSOR_POS_NET_NORM2,                          "posnet.%d.norm2" },
+    { LLM_TENSOR_POS_NET_ATTN_NORM,                      "posnet.%d.attn_norm" },
+    { LLM_TENSOR_POS_NET_ATTN_Q,                         "posnet.%d.attn_q" },
+    { LLM_TENSOR_POS_NET_ATTN_K,                         "posnet.%d.attn_k" },
+    { LLM_TENSOR_POS_NET_ATTN_V,                         "posnet.%d.attn_v" },
+    { LLM_TENSOR_POS_NET_ATTN_OUT,                       "posnet.%d.attn_output" },
+    { LLM_TENSOR_ATTN_SINKS,                             "blk.%d.attn_sinks" },
+    { LLM_TENSOR_SHORTCONV_CONV,                         "blk.%d.shortconv.conv" },
+    { LLM_TENSOR_SHORTCONV_INPROJ,                       "blk.%d.shortconv.in_proj" },
+    { LLM_TENSOR_SHORTCONV_OUTPROJ,                      "blk.%d.shortconv.out_proj" },
+    { LLM_TENSOR_FFN_GATE_CHEXPS,                        "blk.%d.ffn_gate_chexps" },
+    { LLM_TENSOR_FFN_DOWN_CHEXPS,                        "blk.%d.ffn_down_chexps" },
+    { LLM_TENSOR_FFN_UP_CHEXPS,                          "blk.%d.ffn_up_chexps" },
+    { LLM_TENSOR_VISEXP_ATTN_QKV,                        "blk.%d.vis_attn_qkv" },
+    { LLM_TENSOR_VISEXP_ATTN_OUT,                        "blk.%d.vis_attn_output" },
+    { LLM_TENSOR_VISEXP_FFN_GATE,                        "blk.%d.vis_gate" },
+    { LLM_TENSOR_VISEXP_FFN_DOWN,                        "blk.%d.vis_down" },
+    { LLM_TENSOR_VISEXP_FFN_UP,                          "blk.%d.vis_up" },
+    { LLM_TENSOR_INDEXER_K_NORM,                         "blk.%d.indexer.k_norm" },
+    { LLM_TENSOR_INDEXER_PROJ,                           "blk.%d.indexer.proj" },
+    { LLM_TENSOR_INDEXER_ATTN_K,                         "blk.%d.indexer.attn_k" },
+    { LLM_TENSOR_INDEXER_ATTN_Q_B,                       "blk.%d.indexer.attn_q_b" },
+};
+static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
+    switch (arch) {
+        case LLM_ARCH_CLIP:
+            return {};
+        case LLM_ARCH_LLAMA:
+        case LLM_ARCH_DECI:
+        case LLM_ARCH_MISTRAL3:
+        case LLM_ARCH_LLAMA_EMBED:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXP,
+                LLM_TENSOR_FFN_DOWN_EXP,
+                LLM_TENSOR_FFN_UP_EXP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_ARCEE:
+        case LLM_ARCH_STARCODER2:
+        case LLM_ARCH_NEMOTRON:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_AFMOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_GATE,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_LLAMA4:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXP,
+                LLM_TENSOR_FFN_DOWN_EXP,
+                LLM_TENSOR_FFN_UP_EXP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_BAICHUAN:
+        case LLM_ARCH_ORION:
+        case LLM_ARCH_XVERSE:
+        case LLM_ARCH_EXAONE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_FALCON:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_NORM_2,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_GROK:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXP,
+                LLM_TENSOR_FFN_DOWN_EXP,
+                LLM_TENSOR_FFN_UP_EXP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_LAYER_OUT_NORM,
+                LLM_TENSOR_ATTN_OUT_NORM,
+            };
+        case LLM_ARCH_GPT2:
+        case LLM_ARCH_STARCODER:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_POS_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_GPTNEOX:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_MPT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_ACT,
+                LLM_TENSOR_POS_EMBD,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+            };
+        case LLM_ARCH_REFACT:
+        case LLM_ARCH_QWEN2:
+        case LLM_ARCH_QWEN2VL:
+        case LLM_ARCH_INTERNLM2:
+        case LLM_ARCH_GRANITE:
+        case LLM_ARCH_ERNIE4_5:
+        case LLM_ARCH_PADDLEOCR:
+        case LLM_ARCH_SMOLLM3:
+        case LLM_ARCH_DREAM:
+        case LLM_ARCH_LLADA:
+        case LLM_ARCH_PANGU_EMBED:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_BERT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_TOKEN_TYPES,
+                LLM_TENSOR_POS_EMBD,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_LAYER_OUT_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_CLS,
+                LLM_TENSOR_CLS_OUT,
+            };
+        case LLM_ARCH_NOMIC_BERT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_TOKEN_TYPES,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_LAYER_OUT_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_NOMIC_BERT_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_TOKEN_TYPES,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_LAYER_OUT_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_NEO_BERT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_ENC_OUTPUT_NORM,
+                LLM_TENSOR_CLS,
+                LLM_TENSOR_CLS_OUT,
+            };
+        case LLM_ARCH_EUROBERT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_MODERN_BERT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_CLS,
+                LLM_TENSOR_CLS_OUT,
+                LLM_TENSOR_CLS_NORM,
+            };
+        case LLM_ARCH_JINA_BERT_V2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_TOKEN_TYPES,
+                LLM_TENSOR_ATTN_NORM_2,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_LAYER_OUT_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_CLS,
+            };
+        case LLM_ARCH_JINA_BERT_V3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_TOKEN_TYPES,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_LAYER_OUT_NORM,
+            };
+        case LLM_ARCH_BLOOM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_STABLELM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+            };
+        case LLM_ARCH_QWEN:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_QWEN2MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_QWEN3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_CLS_OUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_QWEN3MOE:
+        case LLM_ARCH_QWEN3VLMOE:
+        case LLM_ARCH_OLMOE:
+        case LLM_ARCH_LLADA_MOE:
+        case LLM_ARCH_RND1:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_QWEN3NEXT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_GATE,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_SSM_A_NOSCAN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_BETA_ALPHA,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+            };
+        case LLM_ARCH_QWEN35:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_GATE,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_SSM_A_NOSCAN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_BETA,
+                LLM_TENSOR_SSM_ALPHA,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+            };
+        case LLM_ARCH_QWEN35MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_GATE,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_SSM_A_NOSCAN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_BETA,
+                LLM_TENSOR_SSM_ALPHA,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+            };
+        case LLM_ARCH_QWEN3VL:
+        case LLM_ARCH_CHAMELEON:
+        case LLM_ARCH_HUNYUAN_DENSE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_PHI2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_PHI3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FACTORS_LONG,
+                LLM_TENSOR_ROPE_FACTORS_SHORT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_PHIMOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FACTORS_LONG,
+                LLM_TENSOR_ROPE_FACTORS_SHORT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_PLAMO:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_PLAMO2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_X,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_OUT,
+                LLM_TENSOR_SSM_DT_NORM,
+                LLM_TENSOR_SSM_B_NORM,
+                LLM_TENSOR_SSM_C_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_PLAMO3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_CODESHELL:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_MINICPM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ROPE_FACTORS_LONG,
+                LLM_TENSOR_ROPE_FACTORS_SHORT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXP,
+                LLM_TENSOR_FFN_DOWN_EXP,
+                LLM_TENSOR_FFN_UP_EXP,
+            };
+        case LLM_ARCH_MINICPM3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FACTORS_LONG,
+                LLM_TENSOR_ROPE_FACTORS_SHORT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q_A_NORM,
+                LLM_TENSOR_ATTN_KV_A_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_A,
+                LLM_TENSOR_ATTN_Q_B,
+                LLM_TENSOR_ATTN_KV_A_MQA,
+                LLM_TENSOR_ATTN_KV_B,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_GEMMA:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_GEMMA2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_GEMMA3:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_GEMMA3N:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_PER_LAYER_TOKEN_EMBD,
+                LLM_TENSOR_PER_LAYER_MODEL_PROJ,
+                LLM_TENSOR_PER_LAYER_PROJ_NORM,
+                LLM_TENSOR_ALTUP_UNEMBD_PROJ,
+                LLM_TENSOR_ALTUP_PROJ,
+                LLM_TENSOR_PER_LAYER_INP_GATE,
+                LLM_TENSOR_PER_LAYER_PROJ,
+                LLM_TENSOR_PER_LAYER_POST_NORM,
+                LLM_TENSOR_ALTUP_CORRECT_COEF,
+                LLM_TENSOR_ALTUP_CORRECT_SCALE,
+                LLM_TENSOR_ALTUP_PREDICT_COEF,
+                LLM_TENSOR_ALTUP_ROUTER,
+                LLM_TENSOR_ALTUP_ROUTER_NORM,
+                LLM_TENSOR_LAUREL_L,
+                LLM_TENSOR_LAUREL_R,
+                LLM_TENSOR_LAUREL_POST_NORM,
+            };
+        case LLM_ARCH_GEMMA_EMBEDDING:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_DENSE_2_OUT,
+                LLM_TENSOR_DENSE_3_OUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_MAMBA:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_X,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_OUT,
+            };
+        case LLM_ARCH_MAMBA2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+            };
+        case LLM_ARCH_JAMBA:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_X,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_DT_NORM,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_B_NORM,
+                LLM_TENSOR_SSM_C_NORM,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_OUT,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_FALCON_H1:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_COMMAND_R:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+            };
+        case LLM_ARCH_COHERE2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_DBRX:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_OUT_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_OLMO:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_OLMO2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_OPENELM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_ARCTIC:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_NORM_EXPS,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_DEEPSEEK:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_DEEPSEEK2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q_A_NORM,
+                LLM_TENSOR_ATTN_KV_A_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_A,
+                LLM_TENSOR_ATTN_Q_B,
+                LLM_TENSOR_ATTN_KV_A_MQA,
+                LLM_TENSOR_ATTN_KV_B,
+                LLM_TENSOR_ATTN_K_B,
+                LLM_TENSOR_ATTN_V_B,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_PLM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_KV_A_MQA,
+                LLM_TENSOR_ATTN_KV_A_NORM,
+                LLM_TENSOR_ATTN_KV_B,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_CHATGLM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_GLM4:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_POST_NORM,
+                LLM_TENSOR_NEXTN_EH_PROJ,
+                LLM_TENSOR_NEXTN_EMBED_TOKENS,
+                LLM_TENSOR_NEXTN_ENORM,
+                LLM_TENSOR_NEXTN_HNORM,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
+            };
+        case LLM_ARCH_GLM4_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+                LLM_TENSOR_NEXTN_EH_PROJ,
+                LLM_TENSOR_NEXTN_EMBED_TOKENS,
+                LLM_TENSOR_NEXTN_ENORM,
+                LLM_TENSOR_NEXTN_HNORM,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
+            };
+        case LLM_ARCH_GLM_DSA:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q_A_NORM,
+                LLM_TENSOR_ATTN_KV_A_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_A,
+                LLM_TENSOR_ATTN_Q_B,
+                LLM_TENSOR_ATTN_KV_A_MQA,
+                LLM_TENSOR_ATTN_KV_B,
+                LLM_TENSOR_ATTN_K_B,
+                LLM_TENSOR_ATTN_V_B,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+                LLM_TENSOR_INDEXER_K_NORM,
+                LLM_TENSOR_INDEXER_PROJ,
+                LLM_TENSOR_INDEXER_ATTN_K,
+                LLM_TENSOR_INDEXER_ATTN_Q_B,
+                LLM_TENSOR_NEXTN_EH_PROJ,
+                LLM_TENSOR_NEXTN_EMBED_TOKENS,
+                LLM_TENSOR_NEXTN_ENORM,
+                LLM_TENSOR_NEXTN_HNORM,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
+            };
+        case LLM_ARCH_BITNET:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_SUB_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_SUB_NORM,
+            };
+        case LLM_ARCH_T5:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_DEC_OUTPUT_NORM,
+                LLM_TENSOR_DEC_ATTN_NORM,
+                LLM_TENSOR_DEC_ATTN_Q,
+                LLM_TENSOR_DEC_ATTN_K,
+                LLM_TENSOR_DEC_ATTN_V,
+                LLM_TENSOR_DEC_ATTN_OUT,
+                LLM_TENSOR_DEC_ATTN_REL_B,
+                LLM_TENSOR_DEC_CROSS_ATTN_NORM,
+                LLM_TENSOR_DEC_CROSS_ATTN_Q,
+                LLM_TENSOR_DEC_CROSS_ATTN_K,
+                LLM_TENSOR_DEC_CROSS_ATTN_V,
+                LLM_TENSOR_DEC_CROSS_ATTN_OUT,
+                LLM_TENSOR_DEC_CROSS_ATTN_REL_B,
+                LLM_TENSOR_DEC_FFN_NORM,
+                LLM_TENSOR_DEC_FFN_GATE,
+                LLM_TENSOR_DEC_FFN_DOWN,
+                LLM_TENSOR_DEC_FFN_UP,
+                LLM_TENSOR_ENC_OUTPUT_NORM,
+                LLM_TENSOR_ENC_ATTN_NORM,
+                LLM_TENSOR_ENC_ATTN_Q,
+                LLM_TENSOR_ENC_ATTN_K,
+                LLM_TENSOR_ENC_ATTN_V,
+                LLM_TENSOR_ENC_ATTN_OUT,
+                LLM_TENSOR_ENC_ATTN_REL_B,
+                LLM_TENSOR_ENC_FFN_NORM,
+                LLM_TENSOR_ENC_FFN_GATE,
+                LLM_TENSOR_ENC_FFN_DOWN,
+                LLM_TENSOR_ENC_FFN_UP,
+            };
+        case LLM_ARCH_T5ENCODER:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ENC_OUTPUT_NORM,
+                LLM_TENSOR_ENC_ATTN_NORM,
+                LLM_TENSOR_ENC_ATTN_Q,
+                LLM_TENSOR_ENC_ATTN_K,
+                LLM_TENSOR_ENC_ATTN_V,
+                LLM_TENSOR_ENC_ATTN_OUT,
+                LLM_TENSOR_ENC_ATTN_REL_B,
+                LLM_TENSOR_ENC_FFN_NORM,
+                LLM_TENSOR_ENC_FFN_GATE,
+                LLM_TENSOR_ENC_FFN_DOWN,
+                LLM_TENSOR_ENC_FFN_UP,
+            };
+        case LLM_ARCH_JAIS:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_JAIS2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+            };
+        case LLM_ARCH_NEMOTRON_H:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_NEMOTRON_H_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                // mamba(2) ssm layers
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+                // attention layers
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                // dense FFN
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                // MoE FFN (for MoE layers)
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+                // MoE shared expert layer
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_EXAONE4:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_POST_NORM,
+            };
+        case LLM_ARCH_EXAONE_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+                LLM_TENSOR_NEXTN_EH_PROJ,
+                LLM_TENSOR_NEXTN_EMBED_TOKENS,
+                LLM_TENSOR_NEXTN_ENORM,
+                LLM_TENSOR_NEXTN_HNORM,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
+            };
+        case LLM_ARCH_RWKV6:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_NORM_2,
+                LLM_TENSOR_TIME_MIX_W1,
+                LLM_TENSOR_TIME_MIX_W2,
+                LLM_TENSOR_TIME_MIX_LERP_X,
+                LLM_TENSOR_TIME_MIX_LERP_W,
+                LLM_TENSOR_TIME_MIX_LERP_K,
+                LLM_TENSOR_TIME_MIX_LERP_V,
+                LLM_TENSOR_TIME_MIX_LERP_R,
+                LLM_TENSOR_TIME_MIX_LERP_G,
+                LLM_TENSOR_TIME_MIX_LERP_FUSED,
+                LLM_TENSOR_TIME_MIX_FIRST,
+                LLM_TENSOR_TIME_MIX_DECAY,
+                LLM_TENSOR_TIME_MIX_DECAY_W1,
+                LLM_TENSOR_TIME_MIX_DECAY_W2,
+                LLM_TENSOR_TIME_MIX_KEY,
+                LLM_TENSOR_TIME_MIX_VALUE,
+                LLM_TENSOR_TIME_MIX_RECEPTANCE,
+                LLM_TENSOR_TIME_MIX_GATE,
+                LLM_TENSOR_TIME_MIX_LN,
+                LLM_TENSOR_TIME_MIX_OUTPUT,
+                LLM_TENSOR_CHANNEL_MIX_LERP_K,
+                LLM_TENSOR_CHANNEL_MIX_LERP_R,
+                LLM_TENSOR_CHANNEL_MIX_KEY,
+                LLM_TENSOR_CHANNEL_MIX_VALUE,
+                LLM_TENSOR_CHANNEL_MIX_RECEPTANCE,
+            };
+        case LLM_ARCH_RWKV6QWEN2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_TIME_MIX_W1,
+                LLM_TENSOR_TIME_MIX_W2,
+                LLM_TENSOR_TIME_MIX_LERP_X,
+                LLM_TENSOR_TIME_MIX_LERP_FUSED,
+                LLM_TENSOR_TIME_MIX_FIRST,
+                LLM_TENSOR_TIME_MIX_DECAY,
+                LLM_TENSOR_TIME_MIX_DECAY_W1,
+                LLM_TENSOR_TIME_MIX_DECAY_W2,
+                LLM_TENSOR_TIME_MIX_KEY,
+                LLM_TENSOR_TIME_MIX_VALUE,
+                LLM_TENSOR_TIME_MIX_RECEPTANCE,
+                LLM_TENSOR_TIME_MIX_GATE,
+                LLM_TENSOR_TIME_MIX_OUTPUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_RWKV7:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_NORM_2,
+                LLM_TENSOR_TIME_MIX_W0,
+                LLM_TENSOR_TIME_MIX_W1,
+                LLM_TENSOR_TIME_MIX_W2,
+                LLM_TENSOR_TIME_MIX_A0,
+                LLM_TENSOR_TIME_MIX_A1,
+                LLM_TENSOR_TIME_MIX_A2,
+                LLM_TENSOR_TIME_MIX_V0,
+                LLM_TENSOR_TIME_MIX_V1,
+                LLM_TENSOR_TIME_MIX_V2,
+                LLM_TENSOR_TIME_MIX_G1,
+                LLM_TENSOR_TIME_MIX_G2,
+                LLM_TENSOR_TIME_MIX_K_K,
+                LLM_TENSOR_TIME_MIX_K_A,
+                LLM_TENSOR_TIME_MIX_R_K,
+                LLM_TENSOR_TIME_MIX_LERP_FUSED,
+                LLM_TENSOR_TIME_MIX_KEY,
+                LLM_TENSOR_TIME_MIX_VALUE,
+                LLM_TENSOR_TIME_MIX_RECEPTANCE,
+                LLM_TENSOR_TIME_MIX_LN,
+                LLM_TENSOR_TIME_MIX_OUTPUT,
+                LLM_TENSOR_CHANNEL_MIX_LERP_K,
+                LLM_TENSOR_CHANNEL_MIX_KEY,
+                LLM_TENSOR_CHANNEL_MIX_VALUE,
+            };
+        case LLM_ARCH_ARWKV7:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_TIME_MIX_W0,
+                LLM_TENSOR_TIME_MIX_W1,
+                LLM_TENSOR_TIME_MIX_W2,
+                LLM_TENSOR_TIME_MIX_A0,
+                LLM_TENSOR_TIME_MIX_A1,
+                LLM_TENSOR_TIME_MIX_A2,
+                LLM_TENSOR_TIME_MIX_V0,
+                LLM_TENSOR_TIME_MIX_V1,
+                LLM_TENSOR_TIME_MIX_V2,
+                LLM_TENSOR_TIME_MIX_G1,
+                LLM_TENSOR_TIME_MIX_G2,
+                LLM_TENSOR_TIME_MIX_K_K,
+                LLM_TENSOR_TIME_MIX_K_A,
+                LLM_TENSOR_TIME_MIX_R_K,
+                LLM_TENSOR_TIME_MIX_LERP_FUSED,
+                LLM_TENSOR_TIME_MIX_KEY,
+                LLM_TENSOR_TIME_MIX_VALUE,
+                LLM_TENSOR_TIME_MIX_RECEPTANCE,
+                LLM_TENSOR_TIME_MIX_LN,
+                LLM_TENSOR_TIME_MIX_OUTPUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_GRANITE_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_GRANITE_HYBRID:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_SSM_IN,
+                LLM_TENSOR_SSM_CONV1D,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_D,
+                LLM_TENSOR_SSM_NORM,
+                LLM_TENSOR_SSM_OUT,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_WAVTOKENIZER_DEC:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_CONV1D,
+                LLM_TENSOR_CONVNEXT_DW,
+                LLM_TENSOR_CONVNEXT_NORM,
+                LLM_TENSOR_CONVNEXT_PW1,
+                LLM_TENSOR_CONVNEXT_PW2,
+                LLM_TENSOR_CONVNEXT_GAMMA,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_POS_NET_CONV1,
+                LLM_TENSOR_POS_NET_CONV2,
+                LLM_TENSOR_POS_NET_NORM,
+                LLM_TENSOR_POS_NET_NORM1,
+                LLM_TENSOR_POS_NET_NORM2,
+                LLM_TENSOR_POS_NET_ATTN_NORM,
+                LLM_TENSOR_POS_NET_ATTN_Q,
+                LLM_TENSOR_POS_NET_ATTN_K,
+                LLM_TENSOR_POS_NET_ATTN_V,
+                LLM_TENSOR_POS_NET_ATTN_OUT,
+            };
+        case LLM_ARCH_BAILINGMOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+            };
+        case LLM_ARCH_BAILINGMOE2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_NEXTN_EH_PROJ,
+                LLM_TENSOR_NEXTN_EMBED_TOKENS,
+                LLM_TENSOR_NEXTN_ENORM,
+                LLM_TENSOR_NEXTN_HNORM,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
+                LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
+                LLM_TENSOR_LAYER_OUT_NORM,
+            };
+        case LLM_ARCH_DOTS1:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_INP_SHEXP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_ERNIE4_5_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_HUNYUAN_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_OPENAI_MOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_SINKS,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_LFM2:
+            return {
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_SHORTCONV_CONV,
+                LLM_TENSOR_SHORTCONV_INPROJ,
+                LLM_TENSOR_SHORTCONV_OUTPROJ,
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM_LFM2,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_DENSE_2_OUT,
+            };
+        case LLM_ARCH_LFM2MOE:
+            return {
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_SHORTCONV_CONV,
+                LLM_TENSOR_SHORTCONV_INPROJ,
+                LLM_TENSOR_SHORTCONV_OUTPROJ,
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM_LFM2,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_SMALLTHINKER:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+            };
+        case LLM_ARCH_APERTUS:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_SEED_OSS:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_POST_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_GROVEMOE:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_CHEXPS,
+                LLM_TENSOR_FFN_DOWN_CHEXPS,
+                LLM_TENSOR_FFN_UP_CHEXPS,
+            };
+        case LLM_ARCH_MINIMAX_M2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_COGVLM:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_VISEXP_ATTN_QKV,
+                LLM_TENSOR_VISEXP_ATTN_OUT,
+                LLM_TENSOR_VISEXP_FFN_GATE,
+                LLM_TENSOR_VISEXP_FFN_DOWN,
+                LLM_TENSOR_VISEXP_FFN_UP,
+            };
+        case LLM_ARCH_MIMO2:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_SINKS,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_STEP35:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ROPE_FACTORS_LONG,
+                LLM_TENSOR_ROPE_FACTORS_SHORT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_GATE,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+            };
+        case LLM_ARCH_GPTJ:
+        case LLM_ARCH_UNKNOWN:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+            };
+        case LLM_ARCH_MAINCODER:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_Q_NORM,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_K_NORM,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+            };
+        case LLM_ARCH_KIMI_LINEAR:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_OUTPUT,
+                LLM_TENSOR_ROPE_FREQS,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_Q,
+                LLM_TENSOR_ATTN_K,
+                LLM_TENSOR_ATTN_V,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_FFN_NORM,
+                // Dense FFN (layer 0 only)
+                LLM_TENSOR_FFN_GATE,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                // MoE FFN (layers 1+)
+                LLM_TENSOR_FFN_GATE_INP,
+                LLM_TENSOR_FFN_GATE_EXPS,
+                LLM_TENSOR_FFN_DOWN_EXPS,
+                LLM_TENSOR_FFN_UP_EXPS,
+                LLM_TENSOR_FFN_EXP_PROBS_B,
+                // Shared experts
+                LLM_TENSOR_FFN_GATE_SHEXP,
+                LLM_TENSOR_FFN_DOWN_SHEXP,
+                LLM_TENSOR_FFN_UP_SHEXP,
+                // KDA (using SSM_ enum prefix, keeping GGUF names for backward compat)
+                LLM_TENSOR_SSM_CONV1D_Q,
+                LLM_TENSOR_SSM_CONV1D_K,
+                LLM_TENSOR_SSM_CONV1D_V,
+                LLM_TENSOR_SSM_F_A,
+                LLM_TENSOR_SSM_F_B,
+                LLM_TENSOR_SSM_BETA,
+                LLM_TENSOR_SSM_A,
+                LLM_TENSOR_SSM_G_A,
+                LLM_TENSOR_SSM_G_B,
+                LLM_TENSOR_SSM_DT,
+                LLM_TENSOR_SSM_NORM,
+                // MLA
+                LLM_TENSOR_ATTN_Q_A,
+                LLM_TENSOR_ATTN_Q_B,
+                LLM_TENSOR_ATTN_Q_A_NORM,
+                LLM_TENSOR_ATTN_KV_A_MQA,
+                LLM_TENSOR_ATTN_KV_B,
+                LLM_TENSOR_ATTN_K_B,
+                LLM_TENSOR_ATTN_V_B,
+                LLM_TENSOR_ATTN_KV_A_NORM,
+            };
+        default:
+            GGML_ABORT("unknown architecture for tensor mapping");
+    }
+}
+// declare information about the model weight tensors:
+// - the layer in which the tensor is going to be used. this is needed in order to assign the correct buffer type for the weight
+// - the operator which is going to use the weight. this is needed to determine if the respective backend supports the operator
+//
+// for example, input layers are usually assigned to CPU/host buffer types
+//
+// a mismatch between the declared information and the actual layer/op in which the tensor is used can lead to sub-optimal
+//   assignment of the buffer types and extra overhead during computation
+// example: https://github.com/ggml-org/llama.cpp/pull/17548
+//
+static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
+    {LLM_TENSOR_TOKEN_EMBD,                 {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
+    {LLM_TENSOR_POS_EMBD,                   {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
+    {LLM_TENSOR_TOKEN_TYPES,                {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
+    {LLM_TENSOR_TOKEN_EMBD_NORM,            {LLM_TENSOR_LAYER_INPUT, GGML_OP_MUL}},
+    {LLM_TENSOR_OUTPUT,                     {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_CLS,                        {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_CLS_OUT,                    {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_CLS_NORM,                   {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
+    {LLM_TENSOR_DENSE_2_OUT,                {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}}, // Dense layer output
+    {LLM_TENSOR_DENSE_3_OUT,                {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}}, // Dense layer output
+    {LLM_TENSOR_OUTPUT_NORM,                {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
+    {LLM_TENSOR_OUTPUT_NORM_LFM2,           {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
+    {LLM_TENSOR_DEC_OUTPUT_NORM,            {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
+    {LLM_TENSOR_ENC_OUTPUT_NORM,            {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
+    {LLM_TENSOR_ROPE_FREQS,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ROPE}},
+    {LLM_TENSOR_ROPE_FACTORS_LONG,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ROPE}},
+    {LLM_TENSOR_ROPE_FACTORS_SHORT,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ROPE}},
+    {LLM_TENSOR_ATTN_Q,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_K,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_V,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_QKV,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_OUT,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_GATE,                  {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_FFN_GATE,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_FFN_DOWN,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_FFN_UP,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_FFN_DOWN_SHEXP,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_FFN_GATE_SHEXP,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_FFN_UP_SHEXP,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_Q_A,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_Q_B,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_KV_A_MQA,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_KV_B,                  {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_K_B,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_V_B,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ATTN_SINKS,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SCALE}},
+    {LLM_TENSOR_DEC_ATTN_Q,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_DEC_ATTN_K,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_DEC_ATTN_V,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_DEC_ATTN_OUT,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_DEC_CROSS_ATTN_Q,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_DEC_CROSS_ATTN_K,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_DEC_CROSS_ATTN_V,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_DEC_CROSS_ATTN_OUT,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_DEC_FFN_GATE,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_DEC_FFN_DOWN,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_DEC_FFN_UP,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ENC_ATTN_Q,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ENC_ATTN_K,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ENC_ATTN_V,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ENC_ATTN_OUT,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ENC_FFN_GATE,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ENC_FFN_DOWN,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ENC_FFN_UP,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_FFN_GATE_INP_SHEXP,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_FFN_GATE_INP,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SSM_IN,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SSM_X,                      {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SSM_DT,                     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SSM_OUT,                    {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SSM_ALPHA,                  {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SSM_BETA_ALPHA,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_W1,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_W2,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_A1,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_A2,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_V1,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_V2,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_G1,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_G2,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_DECAY_W1,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_DECAY_W2,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_KEY,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_VALUE,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_RECEPTANCE,        {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_GATE,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_OUTPUT,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_CHANNEL_MIX_KEY,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_CHANNEL_MIX_RECEPTANCE,     {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_CHANNEL_MIX_VALUE,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_FFN_ACT,                    {LLM_TENSOR_LAYER_REPEATING, GGML_OP_DIV}},
+    {LLM_TENSOR_SSM_CONV1D,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SSM_CONV}},
+    {LLM_TENSOR_SSM_A,                      {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SSM_SCAN}},
+    {LLM_TENSOR_SSM_A_NOSCAN,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}}, // a version of SSM_A used for MUL instead of SSM_SCAN
+    {LLM_TENSOR_SSM_DT_NORM,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_SSM_B_NORM,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_SSM_C_NORM,                 {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_SSM_D,                      {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_SSM_NORM,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    // Kimi KDA - Conv tensors are 4D [d_conv, 1, d_inner, 1], reshaped to 2D at runtime
+    {LLM_TENSOR_SSM_CONV1D_Q,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_SSM_CONV1D_K,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_SSM_CONV1D_V,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_SSM_F_A,                    {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SSM_F_B,                    {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SSM_BETA,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SSM_G_A,                    {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SSM_G_B,                    {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_TIME_MIX_LERP_X,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_TIME_MIX_LN,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_CHANNEL_MIX_LERP_K,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_CHANNEL_MIX_LERP_R,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_TIME_MIX_K_K,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_TIME_MIX_K_A,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_TIME_MIX_R_K,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_TIME_MIX_LERP_W,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    {LLM_TENSOR_TIME_MIX_LERP_K,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    {LLM_TENSOR_TIME_MIX_LERP_V,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    {LLM_TENSOR_TIME_MIX_LERP_R,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    {LLM_TENSOR_TIME_MIX_LERP_G,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    {LLM_TENSOR_TIME_MIX_LERP_FUSED,        {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    {LLM_TENSOR_TIME_MIX_DECAY,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    {LLM_TENSOR_TIME_MIX_W0,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    {LLM_TENSOR_TIME_MIX_A0,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    {LLM_TENSOR_TIME_MIX_V0,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    {LLM_TENSOR_TIME_MIX_FIRST,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_RWKV_WKV6}},
+    {LLM_TENSOR_ATTN_NORM,                  {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ATTN_NORM_2,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ATTN_OUT_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ATTN_POST_NORM,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_FFN_NORM,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_FFN_POST_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_FFN_NORM_EXPS,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ATTN_Q_NORM,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ATTN_K_NORM,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_LAYER_OUT_NORM,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ATTN_Q_A_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ATTN_KV_A_NORM,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ATTN_SUB_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_FFN_SUB_NORM,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_DEC_ATTN_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_DEC_CROSS_ATTN_NORM,        {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_DEC_FFN_NORM,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ENC_ATTN_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ENC_FFN_NORM,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_DEC_ATTN_REL_B,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_GET_ROWS}},
+    {LLM_TENSOR_ENC_ATTN_REL_B,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_GET_ROWS}},
+    {LLM_TENSOR_FFN_DOWN_EXPS,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
+    {LLM_TENSOR_FFN_GATE_EXPS,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
+    {LLM_TENSOR_FFN_UP_EXPS,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
+    {LLM_TENSOR_FFN_GATE_UP_EXPS,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
+    {LLM_TENSOR_FFN_DOWN_CHEXPS,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
+    {LLM_TENSOR_FFN_GATE_CHEXPS,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
+    {LLM_TENSOR_FFN_UP_CHEXPS,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
+    {LLM_TENSOR_FFN_EXP_PROBS_B,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    // altup / laurel (gemma 3n)
+    {LLM_TENSOR_PER_LAYER_TOKEN_EMBD,       {LLM_TENSOR_LAYER_OUTPUT,    GGML_OP_GET_ROWS}},
+    {LLM_TENSOR_PER_LAYER_MODEL_PROJ,       {LLM_TENSOR_LAYER_OUTPUT,    GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_PER_LAYER_PROJ_NORM,        {LLM_TENSOR_LAYER_OUTPUT,    GGML_OP_MUL}},
+    {LLM_TENSOR_ALTUP_PROJ,                 {LLM_TENSOR_LAYER_OUTPUT,    GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ALTUP_UNEMBD_PROJ,          {LLM_TENSOR_LAYER_OUTPUT,    GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_PER_LAYER_INP_GATE,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_PER_LAYER_PROJ,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_PER_LAYER_POST_NORM,        {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ALTUP_CORRECT_COEF,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ALTUP_CORRECT_SCALE,        {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ALTUP_PREDICT_COEF,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ALTUP_ROUTER,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ALTUP_ROUTER_NORM,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_LAUREL_L,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_LAUREL_R,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_LAUREL_POST_NORM,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    // this tensor is loaded for T5, but never used
+    {LLM_TENSOR_DEC_CROSS_ATTN_REL_B,       {LLM_TENSOR_LAYER_REPEATING, GGML_OP_NONE}},
+    {LLM_TENSOR_CONV1D,                     {LLM_TENSOR_LAYER_INPUT,     GGML_OP_IM2COL}},
+    {LLM_TENSOR_POS_NET_NORM,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_POS_NET_NORM1,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_POS_NET_NORM2,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_POS_NET_CONV1,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_IM2COL}},
+    {LLM_TENSOR_POS_NET_CONV2,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_IM2COL}},
+    {LLM_TENSOR_POS_NET_ATTN_NORM,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_POS_NET_ATTN_Q,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_POS_NET_ATTN_K,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_POS_NET_ATTN_V,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_POS_NET_ATTN_OUT,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_CONVNEXT_DW,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_IM2COL}},
+    {LLM_TENSOR_CONVNEXT_NORM,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_CONVNEXT_PW1,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_CONVNEXT_PW2,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_CONVNEXT_GAMMA,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_SHORTCONV_CONV,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SSM_CONV}},
+    {LLM_TENSOR_SHORTCONV_INPROJ,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_SHORTCONV_OUTPROJ,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_VISEXP_ATTN_QKV,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_VISEXP_ATTN_OUT,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_VISEXP_FFN_GATE,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_VISEXP_FFN_DOWN,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_VISEXP_FFN_UP,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_INDEXER_K_NORM,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_INDEXER_PROJ,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_INDEXER_ATTN_K,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_INDEXER_ATTN_Q_B,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    // NextN/MTP tensors are currently ignored (reserved for future MTP support)
+    // These tensors only exist in the last layer(s) and are treated as output tensors
+    {LLM_TENSOR_NEXTN_EH_PROJ,              {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_NEXTN_EMBED_TOKENS,         {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_GET_ROWS}},
+    {LLM_TENSOR_NEXTN_ENORM,                {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_GET_ROWS}},
+    {LLM_TENSOR_NEXTN_HNORM,                {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
+    {LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,     {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,     {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
+};
+LLM_KV::LLM_KV(llm_arch arch, const char * suffix) : arch(arch), suffix(suffix) {}
+std::string LLM_KV::operator()(llm_kv kv) const {
+    std::string name = ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch));
+    if (suffix != nullptr) {
+        name += ".";
+        name += suffix;
+    }
+    return name;
+}
+LLM_TN_IMPL::LLM_TN_IMPL(llm_arch arch, llm_tensor tensor, const char * suffix, int bid, int xid)
+    : arch(arch), tensor(tensor), suffix(suffix), bid(bid), xid(xid),
+      model_tensors(llm_get_tensor_names(arch)) {}
+std::string LLM_TN_IMPL::str() const {
+    if (LLM_TENSOR_NAMES.find(tensor) == LLM_TENSOR_NAMES.end()) {
+        GGML_ABORT("unknown tensor name for tensor id %d", static_cast<int>(tensor));
+    }
+    if (model_tensors.find(tensor) == model_tensors.end()) {
+        return LLM_TENSOR_NAMES.at(tensor);
+    }
+    std::string name = ::format(LLM_TENSOR_NAMES.at(tensor), bid, xid);
+    if (suffix != nullptr) {
+        name += ".";
+        name += suffix;
+    }
+    return name;
+}
+const char * llm_arch_name(llm_arch arch) {
+    auto it = LLM_ARCH_NAMES.find(arch);
+    if (it == LLM_ARCH_NAMES.end()) {
+        return "unknown";
+    }
+    return it->second;
+}
+llm_arch llm_arch_from_string(const std::string & name) {
+    for (const auto & kv : LLM_ARCH_NAMES) { // NOLINT
+        if (kv.second == name) {
+            return kv.first;
+        }
+    }
+    return LLM_ARCH_UNKNOWN;
+}
+const llm_tensor_info & llm_tensor_info_for(llm_tensor tensor) {
+    return LLM_TENSOR_INFOS.at(tensor);
+}
+bool llm_arch_is_recurrent(const llm_arch & arch) {
+    switch (arch) {
+        case LLM_ARCH_MAMBA:
+        case LLM_ARCH_MAMBA2:
+        case LLM_ARCH_RWKV6:
+        case LLM_ARCH_RWKV6QWEN2:
+        case LLM_ARCH_RWKV7:
+        case LLM_ARCH_ARWKV7:
+            return true;
+        default:
+            return false;
+    }
+}
+bool llm_arch_is_hybrid(const llm_arch & arch) {
+    switch (arch) {
+        case LLM_ARCH_JAMBA:
+        case LLM_ARCH_FALCON_H1:
+        case LLM_ARCH_PLAMO2:
+        case LLM_ARCH_GRANITE_HYBRID:
+        case LLM_ARCH_LFM2:
+        case LLM_ARCH_LFM2MOE:
+        case LLM_ARCH_NEMOTRON_H:
+        case LLM_ARCH_NEMOTRON_H_MOE:
+        case LLM_ARCH_QWEN3NEXT:
+        case LLM_ARCH_KIMI_LINEAR:
+        case LLM_ARCH_QWEN35:
+        case LLM_ARCH_QWEN35MOE:
+            return true;
+        default:
+            return false;
+    }
+}
+bool llm_arch_is_diffusion(const llm_arch & arch) {
+    switch (arch) {
+        case LLM_ARCH_DREAM:
+        case LLM_ARCH_LLADA:
+        case LLM_ARCH_LLADA_MOE:
+        case LLM_ARCH_RND1:
+            return true;
+        default:
+            return false;
+    }
+}