whispercpp 1.3.2 → 1.3.3
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.
- checksums.yaml +4 -4
- data/.gitignore +6 -3
- data/README.md +71 -14
- data/Rakefile +20 -7
- data/ext/.gitignore +4 -6
- data/ext/dependencies.rb +36 -24
- data/ext/extconf.rb +1 -1
- data/ext/options.rb +48 -184
- data/ext/ruby_whisper.c +18 -0
- data/ext/ruby_whisper_context.c +43 -12
- data/ext/ruby_whisper_model.c +1 -1
- data/ext/ruby_whisper_params.c +4 -2
- data/ext/ruby_whisper_segment.c +81 -4
- data/ext/ruby_whisper_transcribe.cpp +13 -7
- data/ext/ruby_whisper_vad_params.c +1 -1
- data/ext/sources/CMakeLists.txt +5 -1
- data/ext/sources/bindings/javascript/package.json +1 -1
- data/ext/sources/examples/addon.node/__test__/whisper.spec.js +120 -24
- data/ext/sources/examples/addon.node/addon.cpp +150 -31
- data/ext/sources/examples/addon.node/index.js +3 -0
- data/ext/sources/examples/addon.node/vad-example.js +132 -0
- data/ext/sources/examples/bench/bench.cpp +3 -2
- data/ext/sources/examples/cli/cli.cpp +3 -2
- data/ext/sources/examples/command/command.cpp +32 -8
- data/ext/sources/examples/common-whisper.cpp +14 -7
- data/ext/sources/examples/lsp/lsp.cpp +2 -0
- data/ext/sources/examples/quantize/quantize.cpp +3 -0
- data/ext/sources/examples/server/CMakeLists.txt +3 -0
- data/ext/sources/examples/server/server.cpp +169 -22
- data/ext/sources/examples/stream/stream.cpp +6 -0
- data/ext/sources/examples/talk-llama/CMakeLists.txt +4 -1
- data/ext/sources/examples/talk-llama/llama-arch.cpp +171 -3
- data/ext/sources/examples/talk-llama/llama-arch.h +28 -1
- data/ext/sources/examples/talk-llama/llama-batch.cpp +741 -272
- data/ext/sources/examples/talk-llama/llama-batch.h +112 -54
- data/ext/sources/examples/talk-llama/llama-chat.cpp +30 -8
- data/ext/sources/examples/talk-llama/llama-chat.h +1 -0
- data/ext/sources/examples/talk-llama/llama-context.cpp +520 -351
- data/ext/sources/examples/talk-llama/llama-context.h +38 -17
- data/ext/sources/examples/talk-llama/llama-cparams.cpp +1 -1
- data/ext/sources/examples/talk-llama/llama-cparams.h +1 -1
- data/ext/sources/examples/talk-llama/llama-graph.cpp +447 -372
- data/ext/sources/examples/talk-llama/llama-graph.h +128 -58
- data/ext/sources/examples/talk-llama/llama-hparams.cpp +10 -2
- data/ext/sources/examples/talk-llama/llama-hparams.h +19 -2
- data/ext/sources/examples/talk-llama/llama-kv-cache-unified-iswa.cpp +279 -0
- data/ext/sources/examples/talk-llama/llama-kv-cache-unified-iswa.h +128 -0
- data/ext/sources/examples/talk-llama/llama-kv-cache-unified.cpp +1841 -0
- data/ext/sources/examples/talk-llama/llama-kv-cache-unified.h +303 -0
- data/ext/sources/examples/talk-llama/llama-kv-cache.h +14 -472
- data/ext/sources/examples/talk-llama/llama-kv-cells.h +86 -26
- data/ext/sources/examples/talk-llama/llama-memory-hybrid.cpp +246 -0
- data/ext/sources/examples/talk-llama/llama-memory-hybrid.h +138 -0
- data/ext/sources/examples/talk-llama/llama-memory-recurrent.cpp +1125 -0
- data/ext/sources/examples/talk-llama/llama-memory-recurrent.h +183 -0
- data/ext/sources/examples/talk-llama/llama-memory.cpp +58 -0
- data/ext/sources/examples/talk-llama/llama-memory.h +88 -4
- data/ext/sources/examples/talk-llama/llama-mmap.cpp +1 -1
- data/ext/sources/examples/talk-llama/llama-model-loader.cpp +42 -17
- data/ext/sources/examples/talk-llama/llama-model-saver.cpp +1 -0
- data/ext/sources/examples/talk-llama/llama-model.cpp +1863 -563
- data/ext/sources/examples/talk-llama/llama-model.h +27 -0
- data/ext/sources/examples/talk-llama/llama-quant.cpp +89 -6
- data/ext/sources/examples/talk-llama/llama-vocab.cpp +65 -28
- data/ext/sources/examples/talk-llama/llama-vocab.h +1 -0
- data/ext/sources/examples/talk-llama/llama.cpp +11 -7
- data/ext/sources/examples/talk-llama/llama.h +147 -40
- data/ext/sources/examples/talk-llama/talk-llama.cpp +2 -0
- data/ext/sources/examples/talk-llama/unicode.cpp +5 -0
- data/ext/sources/examples/vad-speech-segments/speech.cpp +6 -0
- data/ext/sources/examples/wchess/wchess.cmd/wchess.cmd.cpp +2 -0
- data/ext/sources/ggml/CMakeLists.txt +48 -3
- data/ext/sources/ggml/cmake/common.cmake +24 -0
- data/ext/sources/ggml/include/ggml-backend.h +1 -1
- data/ext/sources/ggml/include/ggml-cpu.h +2 -0
- data/ext/sources/ggml/include/ggml.h +144 -5
- data/ext/sources/ggml/src/CMakeLists.txt +82 -24
- data/ext/sources/ggml/src/ggml-backend-reg.cpp +5 -0
- data/ext/sources/ggml/src/ggml-backend.cpp +46 -23
- data/ext/sources/ggml/src/ggml-blas/CMakeLists.txt +3 -3
- data/ext/sources/ggml/src/ggml-cann/CMakeLists.txt +1 -0
- data/ext/sources/ggml/src/ggml-cann/common.h +6 -1
- data/ext/sources/ggml/src/ggml-cann/ggml-cann.cpp +33 -9
- data/ext/sources/ggml/src/ggml-common.h +4 -0
- data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +133 -40
- data/ext/sources/ggml/src/ggml-cpu/amx/amx.cpp +1 -1
- data/ext/sources/ggml/src/ggml-cpu/amx/mmq.cpp +11 -10
- data/ext/sources/ggml/src/ggml-cpu/arch/arm/cpu-feats.cpp +94 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c +4114 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/arm/repack.cpp +2163 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/loongarch/quants.c +2639 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/powerpc/cpu-feats.cpp +82 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/powerpc/quants.c +2732 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/riscv/quants.c +2069 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/riscv/repack.cpp +397 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/s390/quants.c +1300 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/wasm/quants.c +1481 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/x86/quants.c +4311 -0
- data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-aarch64.cpp → arch/x86/repack.cpp} +79 -3225
- data/ext/sources/ggml/src/ggml-cpu/arch-fallback.h +184 -0
- data/ext/sources/ggml/src/ggml-cpu/common.h +4 -3
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-impl.h +16 -7
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +146 -105
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.cpp +12 -8
- data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-hbm.cpp → hbm.cpp} +1 -1
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +1 -1
- data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +58 -8
- data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.h +5 -0
- data/ext/sources/ggml/src/ggml-cpu/ops.cpp +1057 -174
- data/ext/sources/ggml/src/ggml-cpu/ops.h +8 -0
- data/ext/sources/ggml/src/ggml-cpu/quants.c +1158 -0
- data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-quants.h → quants.h} +26 -0
- data/ext/sources/ggml/src/ggml-cpu/repack.cpp +1571 -0
- data/ext/sources/ggml/src/ggml-cpu/repack.h +98 -0
- data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +330 -38
- data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-traits.cpp → traits.cpp} +1 -1
- data/ext/sources/ggml/src/ggml-cpu/vec.cpp +111 -18
- data/ext/sources/ggml/src/ggml-cpu/vec.h +303 -94
- data/ext/sources/ggml/src/ggml-cuda/common.cuh +60 -37
- data/ext/sources/ggml/src/ggml-cuda/conv2d-dw.cu +161 -0
- data/ext/sources/ggml/src/ggml-cuda/conv2d-dw.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/conv2d-transpose.cu +91 -0
- data/ext/sources/ggml/src/ggml-cuda/conv2d-transpose.cuh +4 -0
- data/ext/sources/ggml/src/ggml-cuda/convert.cu +22 -0
- data/ext/sources/ggml/src/ggml-cuda/convert.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +2 -2
- data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +5 -2
- data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +4 -0
- data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +265 -123
- data/ext/sources/ggml/src/ggml-cuda/mean.cu +19 -0
- data/ext/sources/ggml/src/ggml-cuda/mean.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/mmv.cu +257 -87
- data/ext/sources/ggml/src/ggml-cuda/mmv.cuh +2 -3
- data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cu +6 -4
- data/ext/sources/ggml/src/ggml-cuda/sumrows.cu +5 -18
- data/ext/sources/ggml/src/ggml-cuda/sumrows.cuh +0 -1
- data/ext/sources/ggml/src/ggml-cuda/unary.cu +89 -0
- data/ext/sources/ggml/src/ggml-cuda/unary.cuh +7 -0
- data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +4 -0
- data/ext/sources/ggml/src/ggml-impl.h +127 -183
- data/ext/sources/ggml/src/ggml-metal/CMakeLists.txt +11 -10
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +27 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal.m +331 -49
- data/ext/sources/ggml/src/ggml-metal/ggml-metal.metal +564 -282
- data/ext/sources/ggml/src/ggml-musa/mudnn.cuh +2 -2
- data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +14 -0
- data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +1859 -489
- data/ext/sources/ggml/src/ggml-opencl/kernels/argsort.cl +86 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/concat.cl +109 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/div.cl +72 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/glu.cl +201 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/group_norm.cl +72 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_q4_0_f32_8x_flat.cl +283 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/pad.cl +30 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/repeat.cl +39 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/sigmoid.cl +29 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/sub.cl +72 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/sum_rows.cl +39 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/tanh.cl +63 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/tsembd.cl +48 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/upscale.cl +121 -0
- data/ext/sources/ggml/src/ggml-quants.c +6 -8
- data/ext/sources/ggml/src/ggml-rpc/ggml-rpc.cpp +18 -15
- data/ext/sources/ggml/src/ggml-sycl/CMakeLists.txt +3 -3
- data/ext/sources/ggml/src/ggml-sycl/binbcast.cpp +5 -6
- data/ext/sources/ggml/src/ggml-sycl/common.hpp +20 -48
- data/ext/sources/ggml/src/ggml-sycl/concat.cpp +28 -41
- data/ext/sources/ggml/src/ggml-sycl/conv.cpp +4 -10
- data/ext/sources/ggml/src/ggml-sycl/convert.cpp +117 -165
- data/ext/sources/ggml/src/ggml-sycl/cpy.cpp +192 -53
- data/ext/sources/ggml/src/ggml-sycl/dequantize.hpp +32 -0
- data/ext/sources/ggml/src/ggml-sycl/dmmv.cpp +49 -67
- data/ext/sources/ggml/src/ggml-sycl/dpct/helper.hpp +31 -1
- data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +648 -1039
- data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +18 -9
- data/ext/sources/ggml/src/ggml-sycl/gemm.hpp +3 -0
- data/ext/sources/ggml/src/ggml-sycl/getrows.cpp +8 -105
- data/ext/sources/ggml/src/ggml-sycl/ggml-sycl.cpp +238 -100
- data/ext/sources/ggml/src/ggml-sycl/gla.cpp +2 -2
- data/ext/sources/ggml/src/ggml-sycl/im2col.cpp +1 -1
- data/ext/sources/ggml/src/ggml-sycl/mmq.cpp +60 -80
- data/ext/sources/ggml/src/ggml-sycl/mmvq.cpp +158 -203
- data/ext/sources/ggml/src/ggml-sycl/norm.cpp +55 -74
- data/ext/sources/ggml/src/ggml-sycl/quants.hpp +38 -10
- data/ext/sources/ggml/src/ggml-sycl/rope.cpp +138 -27
- data/ext/sources/ggml/src/ggml-sycl/softmax.cpp +3 -3
- data/ext/sources/ggml/src/ggml-sycl/sycl_hw.cpp +3 -1
- data/ext/sources/ggml/src/ggml-sycl/sycl_hw.hpp +3 -0
- data/ext/sources/ggml/src/ggml-sycl/tsembd.cpp +3 -8
- data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +108 -16
- data/ext/sources/ggml/src/ggml-sycl/wkv.cpp +12 -16
- data/ext/sources/ggml/src/ggml-vulkan/CMakeLists.txt +36 -32
- data/ext/sources/ggml/src/ggml-vulkan/ggml-vulkan.cpp +726 -282
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +4 -12
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv_transpose_1d.comp +98 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu.comp +13 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.comp +15 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.comp +29 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/reglu.comp +9 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +12 -3
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/swiglu.comp +9 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +10 -1
- data/ext/sources/ggml/src/ggml.c +328 -48
- data/ext/sources/ggml/src/ggml.cpp +26 -0
- data/ext/sources/ggml/src/gguf.cpp +24 -3
- data/ext/sources/include/whisper.h +2 -0
- data/ext/sources/src/CMakeLists.txt +2 -0
- data/ext/sources/src/coreml/whisper-compat.h +10 -0
- data/ext/sources/src/coreml/whisper-compat.m +35 -0
- data/ext/sources/src/coreml/whisper-decoder-impl.m +1 -0
- data/ext/sources/src/coreml/whisper-encoder-impl.m +1 -0
- data/ext/sources/src/whisper.cpp +218 -169
- data/extsources.rb +15 -9
- data/lib/whisper/context.rb +15 -0
- data/lib/whisper/model/uri.rb +56 -1
- data/lib/whisper/segment.rb +58 -0
- data/sig/whisper.rbs +68 -38
- data/{tests → test}/helper.rb +1 -12
- data/{tests → test}/test_model.rb +9 -0
- data/test/test_package.rb +51 -0
- data/test/test_segment.rb +146 -0
- data/{tests → test}/test_whisper.rb +70 -0
- data/whispercpp.gemspec +2 -3
- metadata +91 -43
- data/ext/sources/.dockerignore +0 -3
- data/ext/sources/.github/workflows/bindings-ruby.yml +0 -21
- data/ext/sources/ci/run.sh +0 -336
- data/ext/sources/close-issue.yml +0 -28
- data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +0 -2739
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-aarch64.h +0 -8
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-quants.c +0 -13747
- data/tests/test_package.rb +0 -46
- data/tests/test_segment.rb +0 -74
- /data/ext/sources/ggml/src/ggml-cpu/{cpu-feats-x86.cpp → arch/x86/cpu-feats.cpp} +0 -0
- /data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-hbm.h → hbm.h} +0 -0
- /data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-traits.h → traits.h} +0 -0
- /data/{tests → test}/jfk_reader/.gitignore +0 -0
- /data/{tests → test}/jfk_reader/extconf.rb +0 -0
- /data/{tests → test}/jfk_reader/jfk_reader.c +0 -0
- /data/{tests → test}/test_callback.rb +0 -0
- /data/{tests → test}/test_error.rb +0 -0
- /data/{tests → test}/test_params.rb +0 -0
- /data/{tests → test}/test_vad.rb +0 -0
- /data/{tests → test}/test_vad_params.rb +0 -0
@@ -0,0 +1,128 @@
|
|
1
|
+
#pragma once
|
2
|
+
|
3
|
+
#include "llama-kv-cache-unified.h"
|
4
|
+
|
5
|
+
#include <vector>
|
6
|
+
|
7
|
+
//
|
8
|
+
// llama_kv_cache_unified_iswa
|
9
|
+
//
|
10
|
+
|
11
|
+
// utilizes two instances of llama_kv_cache_unified
|
12
|
+
// the first instance is for the non-SWA layers of the model and the second instance is for the SWA layers
|
13
|
+
|
14
|
+
class llama_kv_cache_unified_iswa : public llama_memory_i {
|
15
|
+
public:
|
16
|
+
llama_kv_cache_unified_iswa(
|
17
|
+
const llama_model & model,
|
18
|
+
ggml_type type_k,
|
19
|
+
ggml_type type_v,
|
20
|
+
bool v_trans,
|
21
|
+
bool offload,
|
22
|
+
bool swa_full,
|
23
|
+
uint32_t kv_size,
|
24
|
+
uint32_t n_seq_max,
|
25
|
+
uint32_t n_ubatch,
|
26
|
+
uint32_t n_pad);
|
27
|
+
|
28
|
+
~llama_kv_cache_unified_iswa() = default;
|
29
|
+
|
30
|
+
//
|
31
|
+
// llama_memory_i
|
32
|
+
//
|
33
|
+
|
34
|
+
llama_memory_context_ptr init_batch(
|
35
|
+
llama_batch_allocr & balloc,
|
36
|
+
uint32_t n_ubatch,
|
37
|
+
bool embd_all) override;
|
38
|
+
|
39
|
+
llama_memory_context_ptr init_full() override;
|
40
|
+
|
41
|
+
llama_memory_context_ptr init_update(llama_context * lctx, bool optimize) override;
|
42
|
+
|
43
|
+
bool get_can_shift() const override;
|
44
|
+
|
45
|
+
void clear(bool data) override;
|
46
|
+
|
47
|
+
bool seq_rm (llama_seq_id seq_id, llama_pos p0, llama_pos p1) override;
|
48
|
+
void seq_cp (llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) override;
|
49
|
+
void seq_keep(llama_seq_id seq_id) override;
|
50
|
+
void seq_add (llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos shift) override;
|
51
|
+
void seq_div (llama_seq_id seq_id, llama_pos p0, llama_pos p1, int d) override;
|
52
|
+
|
53
|
+
llama_pos seq_pos_min(llama_seq_id seq_id) const override;
|
54
|
+
llama_pos seq_pos_max(llama_seq_id seq_id) const override;
|
55
|
+
|
56
|
+
// state write/load
|
57
|
+
|
58
|
+
void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1) const override;
|
59
|
+
void state_read (llama_io_read_i & io, llama_seq_id seq_id = -1) override;
|
60
|
+
|
61
|
+
//
|
62
|
+
// llama_kv_cache_unified_iswa specific API
|
63
|
+
//
|
64
|
+
|
65
|
+
llama_kv_cache_unified * get_base() const;
|
66
|
+
llama_kv_cache_unified * get_swa () const;
|
67
|
+
|
68
|
+
private:
|
69
|
+
const llama_hparams & hparams;
|
70
|
+
|
71
|
+
std::unique_ptr<llama_kv_cache_unified> kv_base;
|
72
|
+
std::unique_ptr<llama_kv_cache_unified> kv_swa;
|
73
|
+
};
|
74
|
+
|
75
|
+
class llama_kv_cache_unified_iswa_context : public llama_memory_context_i {
|
76
|
+
public:
|
77
|
+
// used for errors
|
78
|
+
llama_kv_cache_unified_iswa_context(llama_memory_status status);
|
79
|
+
|
80
|
+
// used to create a full-cache context
|
81
|
+
llama_kv_cache_unified_iswa_context(
|
82
|
+
llama_kv_cache_unified_iswa * kv);
|
83
|
+
|
84
|
+
// used to create an update context
|
85
|
+
llama_kv_cache_unified_iswa_context(
|
86
|
+
llama_kv_cache_unified_iswa * kv,
|
87
|
+
llama_context * lctx,
|
88
|
+
bool optimize);
|
89
|
+
|
90
|
+
// used to create a batch processing context from a batch
|
91
|
+
llama_kv_cache_unified_iswa_context(
|
92
|
+
llama_kv_cache_unified_iswa * kv,
|
93
|
+
std::vector<uint32_t> heads_base,
|
94
|
+
std::vector<uint32_t> heads_swa,
|
95
|
+
std::vector<llama_ubatch> ubatches);
|
96
|
+
|
97
|
+
virtual ~llama_kv_cache_unified_iswa_context();
|
98
|
+
|
99
|
+
//
|
100
|
+
// llama_memory_context_i
|
101
|
+
//
|
102
|
+
|
103
|
+
bool next() override;
|
104
|
+
bool apply() override;
|
105
|
+
|
106
|
+
llama_memory_status get_status() const override;
|
107
|
+
const llama_ubatch & get_ubatch() const override;
|
108
|
+
|
109
|
+
//
|
110
|
+
// llama_kv_cache_unified_iswa_context specific API
|
111
|
+
//
|
112
|
+
|
113
|
+
const llama_kv_cache_unified_context * get_base() const;
|
114
|
+
const llama_kv_cache_unified_context * get_swa() const;
|
115
|
+
|
116
|
+
private:
|
117
|
+
//llama_kv_cache_unified_iswa * kv;
|
118
|
+
|
119
|
+
// the index of the next ubatch to process
|
120
|
+
size_t i_next = 0;
|
121
|
+
|
122
|
+
std::vector<llama_ubatch> ubatches;
|
123
|
+
|
124
|
+
const llama_memory_context_ptr ctx_base;
|
125
|
+
const llama_memory_context_ptr ctx_swa;
|
126
|
+
|
127
|
+
const llama_memory_status status;
|
128
|
+
};
|