vllm-cpu-amxbf16 0.11.2.post2__cp310-cp310-manylinux_2_17_x86_64.whl
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.
- vllm/_C.abi3.so +0 -0
- vllm/__init__.py +225 -0
- vllm/_aiter_ops.py +983 -0
- vllm/_bc_linter.py +54 -0
- vllm/_custom_ops.py +2863 -0
- vllm/_ipex_ops.py +457 -0
- vllm/_version.py +34 -0
- vllm/assets/__init__.py +0 -0
- vllm/assets/audio.py +43 -0
- vllm/assets/base.py +40 -0
- vllm/assets/image.py +59 -0
- vllm/assets/video.py +149 -0
- vllm/attention/__init__.py +18 -0
- vllm/attention/backends/__init__.py +0 -0
- vllm/attention/backends/abstract.py +391 -0
- vllm/attention/backends/registry.py +195 -0
- vllm/attention/backends/utils.py +33 -0
- vllm/attention/layer.py +1052 -0
- vllm/attention/layers/__init__.py +0 -0
- vllm/attention/layers/chunked_local_attention.py +121 -0
- vllm/attention/layers/cross_attention.py +178 -0
- vllm/attention/layers/encoder_only_attention.py +103 -0
- vllm/attention/ops/__init__.py +0 -0
- vllm/attention/ops/chunked_prefill_paged_decode.py +401 -0
- vllm/attention/ops/common.py +414 -0
- vllm/attention/ops/flashmla.py +251 -0
- vllm/attention/ops/merge_attn_states.py +47 -0
- vllm/attention/ops/paged_attn.py +262 -0
- vllm/attention/ops/pallas_kv_cache_update.py +130 -0
- vllm/attention/ops/prefix_prefill.py +814 -0
- vllm/attention/ops/rocm_aiter_paged_attn.py +123 -0
- vllm/attention/ops/triton_decode_attention.py +712 -0
- vllm/attention/ops/triton_merge_attn_states.py +105 -0
- vllm/attention/ops/triton_reshape_and_cache_flash.py +184 -0
- vllm/attention/ops/triton_unified_attention.py +941 -0
- vllm/attention/ops/vit_attn_wrappers.py +178 -0
- vllm/attention/selector.py +231 -0
- vllm/attention/utils/__init__.py +0 -0
- vllm/attention/utils/fa_utils.py +109 -0
- vllm/attention/utils/kv_sharing_utils.py +33 -0
- vllm/attention/utils/kv_transfer_utils.py +60 -0
- vllm/beam_search.py +88 -0
- vllm/benchmarks/__init__.py +0 -0
- vllm/benchmarks/datasets.py +3222 -0
- vllm/benchmarks/latency.py +172 -0
- vllm/benchmarks/lib/__init__.py +3 -0
- vllm/benchmarks/lib/endpoint_request_func.py +777 -0
- vllm/benchmarks/lib/ready_checker.py +72 -0
- vllm/benchmarks/lib/utils.py +79 -0
- vllm/benchmarks/serve.py +1531 -0
- vllm/benchmarks/sweep/__init__.py +0 -0
- vllm/benchmarks/sweep/cli.py +38 -0
- vllm/benchmarks/sweep/param_sweep.py +91 -0
- vllm/benchmarks/sweep/plot.py +580 -0
- vllm/benchmarks/sweep/serve.py +416 -0
- vllm/benchmarks/sweep/serve_sla.py +492 -0
- vllm/benchmarks/sweep/server.py +114 -0
- vllm/benchmarks/sweep/sla_sweep.py +132 -0
- vllm/benchmarks/sweep/utils.py +4 -0
- vllm/benchmarks/throughput.py +799 -0
- vllm/collect_env.py +857 -0
- vllm/compilation/__init__.py +0 -0
- vllm/compilation/activation_quant_fusion.py +209 -0
- vllm/compilation/backends.py +759 -0
- vllm/compilation/base_static_graph.py +57 -0
- vllm/compilation/caching.py +178 -0
- vllm/compilation/collective_fusion.py +1234 -0
- vllm/compilation/compiler_interface.py +639 -0
- vllm/compilation/counter.py +48 -0
- vllm/compilation/cuda_graph.py +208 -0
- vllm/compilation/decorators.py +571 -0
- vllm/compilation/fix_functionalization.py +253 -0
- vllm/compilation/fusion.py +374 -0
- vllm/compilation/fusion_attn.py +359 -0
- vllm/compilation/fx_utils.py +91 -0
- vllm/compilation/inductor_pass.py +133 -0
- vllm/compilation/matcher_utils.py +317 -0
- vllm/compilation/monitor.py +62 -0
- vllm/compilation/noop_elimination.py +134 -0
- vllm/compilation/partition_rules.py +72 -0
- vllm/compilation/pass_manager.py +135 -0
- vllm/compilation/piecewise_backend.py +121 -0
- vllm/compilation/post_cleanup.py +21 -0
- vllm/compilation/qk_norm_rope_fusion.py +238 -0
- vllm/compilation/sequence_parallelism.py +363 -0
- vllm/compilation/torch25_custom_graph_pass.py +44 -0
- vllm/compilation/vllm_inductor_pass.py +173 -0
- vllm/compilation/wrapper.py +238 -0
- vllm/config/__init__.py +102 -0
- vllm/config/cache.py +207 -0
- vllm/config/compilation.py +975 -0
- vllm/config/device.py +75 -0
- vllm/config/ec_transfer.py +110 -0
- vllm/config/kv_events.py +56 -0
- vllm/config/kv_transfer.py +114 -0
- vllm/config/load.py +124 -0
- vllm/config/lora.py +112 -0
- vllm/config/model.py +2162 -0
- vllm/config/multimodal.py +248 -0
- vllm/config/observability.py +123 -0
- vllm/config/parallel.py +655 -0
- vllm/config/pooler.py +122 -0
- vllm/config/scheduler.py +298 -0
- vllm/config/speculative.py +654 -0
- vllm/config/speech_to_text.py +38 -0
- vllm/config/structured_outputs.py +92 -0
- vllm/config/utils.py +178 -0
- vllm/config/vllm.py +1166 -0
- vllm/connections.py +189 -0
- vllm/device_allocator/__init__.py +0 -0
- vllm/device_allocator/cumem.py +327 -0
- vllm/distributed/__init__.py +6 -0
- vllm/distributed/communication_op.py +43 -0
- vllm/distributed/device_communicators/__init__.py +0 -0
- vllm/distributed/device_communicators/all2all.py +490 -0
- vllm/distributed/device_communicators/all_reduce_utils.py +344 -0
- vllm/distributed/device_communicators/base_device_communicator.py +297 -0
- vllm/distributed/device_communicators/cpu_communicator.py +209 -0
- vllm/distributed/device_communicators/cuda_communicator.py +340 -0
- vllm/distributed/device_communicators/cuda_wrapper.py +216 -0
- vllm/distributed/device_communicators/custom_all_reduce.py +326 -0
- vllm/distributed/device_communicators/mnnvl_compat.py +27 -0
- vllm/distributed/device_communicators/pynccl.py +386 -0
- vllm/distributed/device_communicators/pynccl_allocator.py +191 -0
- vllm/distributed/device_communicators/pynccl_wrapper.py +564 -0
- vllm/distributed/device_communicators/quick_all_reduce.py +290 -0
- vllm/distributed/device_communicators/ray_communicator.py +259 -0
- vllm/distributed/device_communicators/shm_broadcast.py +733 -0
- vllm/distributed/device_communicators/shm_object_storage.py +660 -0
- vllm/distributed/device_communicators/symm_mem.py +156 -0
- vllm/distributed/device_communicators/tpu_communicator.py +107 -0
- vllm/distributed/device_communicators/xpu_communicator.py +95 -0
- vllm/distributed/ec_transfer/__init__.py +14 -0
- vllm/distributed/ec_transfer/ec_connector/__init__.py +0 -0
- vllm/distributed/ec_transfer/ec_connector/base.py +247 -0
- vllm/distributed/ec_transfer/ec_connector/factory.py +88 -0
- vllm/distributed/ec_transfer/ec_connector/shared_storage_connector.py +201 -0
- vllm/distributed/ec_transfer/ec_transfer_state.py +42 -0
- vllm/distributed/eplb/__init__.py +8 -0
- vllm/distributed/eplb/eplb_state.py +837 -0
- vllm/distributed/eplb/rebalance_algo.py +260 -0
- vllm/distributed/eplb/rebalance_execute.py +431 -0
- vllm/distributed/kv_events.py +371 -0
- vllm/distributed/kv_transfer/README.md +29 -0
- vllm/distributed/kv_transfer/__init__.py +20 -0
- vllm/distributed/kv_transfer/disagg_prefill_workflow.jpg +0 -0
- vllm/distributed/kv_transfer/kv_connector/__init__.py +0 -0
- vllm/distributed/kv_transfer/kv_connector/base.py +10 -0
- vllm/distributed/kv_transfer/kv_connector/factory.py +192 -0
- vllm/distributed/kv_transfer/kv_connector/utils.py +268 -0
- vllm/distributed/kv_transfer/kv_connector/v1/__init__.py +19 -0
- vllm/distributed/kv_transfer/kv_connector/v1/base.py +546 -0
- vllm/distributed/kv_transfer/kv_connector/v1/decode_bench_connector.py +419 -0
- vllm/distributed/kv_transfer/kv_connector/v1/lmcache_connector.py +216 -0
- vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/__init__.py +18 -0
- vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/multi_process_adapter.py +379 -0
- vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/utils.py +221 -0
- vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/vllm_v1_adapter.py +1411 -0
- vllm/distributed/kv_transfer/kv_connector/v1/lmcache_mp_connector.py +867 -0
- vllm/distributed/kv_transfer/kv_connector/v1/metrics.py +189 -0
- vllm/distributed/kv_transfer/kv_connector/v1/multi_connector.py +454 -0
- vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py +2440 -0
- vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py +504 -0
- vllm/distributed/kv_transfer/kv_connector/v1/p2p/__init__.py +0 -0
- vllm/distributed/kv_transfer/kv_connector/v1/p2p/p2p_nccl_connector.py +531 -0
- vllm/distributed/kv_transfer/kv_connector/v1/p2p/p2p_nccl_engine.py +632 -0
- vllm/distributed/kv_transfer/kv_connector/v1/p2p/tensor_memory_pool.py +273 -0
- vllm/distributed/kv_transfer/kv_connector/v1/shared_storage_connector.py +450 -0
- vllm/distributed/kv_transfer/kv_lookup_buffer/__init__.py +0 -0
- vllm/distributed/kv_transfer/kv_lookup_buffer/base.py +179 -0
- vllm/distributed/kv_transfer/kv_lookup_buffer/mooncake_store.py +164 -0
- vllm/distributed/kv_transfer/kv_lookup_buffer/simple_buffer.py +242 -0
- vllm/distributed/kv_transfer/kv_pipe/__init__.py +0 -0
- vllm/distributed/kv_transfer/kv_pipe/base.py +66 -0
- vllm/distributed/kv_transfer/kv_pipe/mooncake_pipe.py +295 -0
- vllm/distributed/kv_transfer/kv_pipe/pynccl_pipe.py +285 -0
- vllm/distributed/kv_transfer/kv_transfer_state.py +78 -0
- vllm/distributed/parallel_state.py +1759 -0
- vllm/distributed/tpu_distributed_utils.py +188 -0
- vllm/distributed/utils.py +543 -0
- vllm/engine/__init__.py +0 -0
- vllm/engine/arg_utils.py +2144 -0
- vllm/engine/async_llm_engine.py +6 -0
- vllm/engine/llm_engine.py +6 -0
- vllm/engine/protocol.py +170 -0
- vllm/entrypoints/__init__.py +0 -0
- vllm/entrypoints/anthropic/__init__.py +0 -0
- vllm/entrypoints/anthropic/protocol.py +162 -0
- vllm/entrypoints/anthropic/serving_messages.py +460 -0
- vllm/entrypoints/api_server.py +184 -0
- vllm/entrypoints/chat_utils.py +1690 -0
- vllm/entrypoints/cli/__init__.py +13 -0
- vllm/entrypoints/cli/benchmark/__init__.py +0 -0
- vllm/entrypoints/cli/benchmark/base.py +25 -0
- vllm/entrypoints/cli/benchmark/latency.py +21 -0
- vllm/entrypoints/cli/benchmark/main.py +56 -0
- vllm/entrypoints/cli/benchmark/serve.py +21 -0
- vllm/entrypoints/cli/benchmark/sweep.py +21 -0
- vllm/entrypoints/cli/benchmark/throughput.py +21 -0
- vllm/entrypoints/cli/collect_env.py +38 -0
- vllm/entrypoints/cli/main.py +79 -0
- vllm/entrypoints/cli/openai.py +256 -0
- vllm/entrypoints/cli/run_batch.py +68 -0
- vllm/entrypoints/cli/serve.py +249 -0
- vllm/entrypoints/cli/types.py +29 -0
- vllm/entrypoints/constants.py +10 -0
- vllm/entrypoints/context.py +572 -0
- vllm/entrypoints/dynamic_lora.py +57 -0
- vllm/entrypoints/harmony_utils.py +535 -0
- vllm/entrypoints/launcher.py +175 -0
- vllm/entrypoints/llm.py +1768 -0
- vllm/entrypoints/logger.py +84 -0
- vllm/entrypoints/openai/__init__.py +0 -0
- vllm/entrypoints/openai/api_server.py +2096 -0
- vllm/entrypoints/openai/cli_args.py +302 -0
- vllm/entrypoints/openai/orca_metrics.py +120 -0
- vllm/entrypoints/openai/protocol.py +3299 -0
- vllm/entrypoints/openai/run_batch.py +547 -0
- vllm/entrypoints/openai/serving_chat.py +1772 -0
- vllm/entrypoints/openai/serving_classification.py +235 -0
- vllm/entrypoints/openai/serving_completion.py +715 -0
- vllm/entrypoints/openai/serving_embedding.py +695 -0
- vllm/entrypoints/openai/serving_engine.py +1433 -0
- vllm/entrypoints/openai/serving_models.py +304 -0
- vllm/entrypoints/openai/serving_pooling.py +346 -0
- vllm/entrypoints/openai/serving_responses.py +2021 -0
- vllm/entrypoints/openai/serving_score.py +503 -0
- vllm/entrypoints/openai/serving_tokenization.py +203 -0
- vllm/entrypoints/openai/serving_tokens.py +269 -0
- vllm/entrypoints/openai/serving_transcription.py +148 -0
- vllm/entrypoints/openai/speech_to_text.py +405 -0
- vllm/entrypoints/openai/tool_parsers/__init__.py +142 -0
- vllm/entrypoints/openai/tool_parsers/abstract_tool_parser.py +273 -0
- vllm/entrypoints/openai/tool_parsers/deepseekv31_tool_parser.py +390 -0
- vllm/entrypoints/openai/tool_parsers/deepseekv3_tool_parser.py +390 -0
- vllm/entrypoints/openai/tool_parsers/ernie45_tool_parser.py +210 -0
- vllm/entrypoints/openai/tool_parsers/glm4_moe_tool_parser.py +200 -0
- vllm/entrypoints/openai/tool_parsers/granite_20b_fc_tool_parser.py +273 -0
- vllm/entrypoints/openai/tool_parsers/granite_tool_parser.py +253 -0
- vllm/entrypoints/openai/tool_parsers/hermes_tool_parser.py +494 -0
- vllm/entrypoints/openai/tool_parsers/hunyuan_a13b_tool_parser.py +420 -0
- vllm/entrypoints/openai/tool_parsers/internlm2_tool_parser.py +227 -0
- vllm/entrypoints/openai/tool_parsers/jamba_tool_parser.py +323 -0
- vllm/entrypoints/openai/tool_parsers/kimi_k2_tool_parser.py +590 -0
- vllm/entrypoints/openai/tool_parsers/llama4_pythonic_tool_parser.py +341 -0
- vllm/entrypoints/openai/tool_parsers/llama_tool_parser.py +290 -0
- vllm/entrypoints/openai/tool_parsers/longcat_tool_parser.py +37 -0
- vllm/entrypoints/openai/tool_parsers/minimax_m2_tool_parser.py +643 -0
- vllm/entrypoints/openai/tool_parsers/minimax_tool_parser.py +849 -0
- vllm/entrypoints/openai/tool_parsers/mistral_tool_parser.py +390 -0
- vllm/entrypoints/openai/tool_parsers/olmo3_tool_parser.py +366 -0
- vllm/entrypoints/openai/tool_parsers/openai_tool_parser.py +97 -0
- vllm/entrypoints/openai/tool_parsers/phi4mini_tool_parser.py +120 -0
- vllm/entrypoints/openai/tool_parsers/pythonic_tool_parser.py +332 -0
- vllm/entrypoints/openai/tool_parsers/qwen3coder_tool_parser.py +781 -0
- vllm/entrypoints/openai/tool_parsers/qwen3xml_tool_parser.py +1316 -0
- vllm/entrypoints/openai/tool_parsers/seed_oss_tool_parser.py +744 -0
- vllm/entrypoints/openai/tool_parsers/step3_tool_parser.py +303 -0
- vllm/entrypoints/openai/tool_parsers/utils.py +229 -0
- vllm/entrypoints/openai/tool_parsers/xlam_tool_parser.py +556 -0
- vllm/entrypoints/renderer.py +409 -0
- vllm/entrypoints/responses_utils.py +77 -0
- vllm/entrypoints/sagemaker/__init__.py +4 -0
- vllm/entrypoints/sagemaker/routes.py +72 -0
- vllm/entrypoints/score_utils.py +242 -0
- vllm/entrypoints/ssl.py +78 -0
- vllm/entrypoints/tool.py +143 -0
- vllm/entrypoints/tool_server.py +209 -0
- vllm/entrypoints/utils.py +319 -0
- vllm/env_override.py +378 -0
- vllm/envs.py +1659 -0
- vllm/forward_context.py +356 -0
- vllm/inputs/__init__.py +44 -0
- vllm/inputs/data.py +359 -0
- vllm/inputs/parse.py +137 -0
- vllm/inputs/preprocess.py +727 -0
- vllm/logger.py +267 -0
- vllm/logging_utils/__init__.py +10 -0
- vllm/logging_utils/dump_input.py +83 -0
- vllm/logging_utils/formatter.py +77 -0
- vllm/logging_utils/log_time.py +34 -0
- vllm/logits_process.py +121 -0
- vllm/logprobs.py +208 -0
- vllm/lora/__init__.py +0 -0
- vllm/lora/layers/__init__.py +41 -0
- vllm/lora/layers/base.py +67 -0
- vllm/lora/layers/base_linear.py +164 -0
- vllm/lora/layers/column_parallel_linear.py +578 -0
- vllm/lora/layers/fused_moe.py +472 -0
- vllm/lora/layers/logits_processor.py +252 -0
- vllm/lora/layers/replicated_linear.py +70 -0
- vllm/lora/layers/row_parallel_linear.py +181 -0
- vllm/lora/layers/utils.py +65 -0
- vllm/lora/layers/vocal_parallel_embedding.py +166 -0
- vllm/lora/lora_weights.py +198 -0
- vllm/lora/models.py +890 -0
- vllm/lora/ops/__init__.py +0 -0
- vllm/lora/ops/ipex_ops/__init__.py +6 -0
- vllm/lora/ops/ipex_ops/lora_ops.py +57 -0
- vllm/lora/ops/torch_ops/__init__.py +20 -0
- vllm/lora/ops/torch_ops/lora_ops.py +128 -0
- vllm/lora/ops/triton_ops/README_TUNING.md +60 -0
- vllm/lora/ops/triton_ops/__init__.py +21 -0
- vllm/lora/ops/triton_ops/fused_moe_lora_op.py +641 -0
- vllm/lora/ops/triton_ops/kernel_utils.py +340 -0
- vllm/lora/ops/triton_ops/lora_expand_op.py +310 -0
- vllm/lora/ops/triton_ops/lora_kernel_metadata.py +154 -0
- vllm/lora/ops/triton_ops/lora_shrink_op.py +287 -0
- vllm/lora/ops/triton_ops/utils.py +295 -0
- vllm/lora/ops/xla_ops/__init__.py +6 -0
- vllm/lora/ops/xla_ops/lora_ops.py +141 -0
- vllm/lora/peft_helper.py +128 -0
- vllm/lora/punica_wrapper/__init__.py +10 -0
- vllm/lora/punica_wrapper/punica_base.py +492 -0
- vllm/lora/punica_wrapper/punica_cpu.py +351 -0
- vllm/lora/punica_wrapper/punica_gpu.py +411 -0
- vllm/lora/punica_wrapper/punica_selector.py +21 -0
- vllm/lora/punica_wrapper/punica_tpu.py +359 -0
- vllm/lora/punica_wrapper/punica_xpu.py +279 -0
- vllm/lora/punica_wrapper/utils.py +150 -0
- vllm/lora/request.py +100 -0
- vllm/lora/resolver.py +88 -0
- vllm/lora/utils.py +293 -0
- vllm/lora/worker_manager.py +279 -0
- vllm/model_executor/__init__.py +11 -0
- vllm/model_executor/custom_op.py +194 -0
- vllm/model_executor/layers/__init__.py +0 -0
- vllm/model_executor/layers/activation.py +569 -0
- vllm/model_executor/layers/attention_layer_base.py +35 -0
- vllm/model_executor/layers/batch_invariant.py +854 -0
- vllm/model_executor/layers/conv.py +236 -0
- vllm/model_executor/layers/fla/__init__.py +8 -0
- vllm/model_executor/layers/fla/ops/__init__.py +17 -0
- vllm/model_executor/layers/fla/ops/chunk.py +240 -0
- vllm/model_executor/layers/fla/ops/chunk_delta_h.py +344 -0
- vllm/model_executor/layers/fla/ops/chunk_o.py +183 -0
- vllm/model_executor/layers/fla/ops/chunk_scaled_dot_kkt.py +154 -0
- vllm/model_executor/layers/fla/ops/cumsum.py +280 -0
- vllm/model_executor/layers/fla/ops/fused_recurrent.py +390 -0
- vllm/model_executor/layers/fla/ops/index.py +41 -0
- vllm/model_executor/layers/fla/ops/kda.py +1351 -0
- vllm/model_executor/layers/fla/ops/l2norm.py +146 -0
- vllm/model_executor/layers/fla/ops/layernorm_guard.py +396 -0
- vllm/model_executor/layers/fla/ops/op.py +60 -0
- vllm/model_executor/layers/fla/ops/solve_tril.py +556 -0
- vllm/model_executor/layers/fla/ops/utils.py +194 -0
- vllm/model_executor/layers/fla/ops/wy_fast.py +158 -0
- vllm/model_executor/layers/fused_moe/__init__.py +106 -0
- vllm/model_executor/layers/fused_moe/all2all_utils.py +160 -0
- vllm/model_executor/layers/fused_moe/batched_deep_gemm_moe.py +406 -0
- vllm/model_executor/layers/fused_moe/batched_triton_or_deep_gemm_moe.py +180 -0
- vllm/model_executor/layers/fused_moe/config.py +916 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=1792,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_H200,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=1,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=NVIDIA_H100,dtype=fp8_w8a8.json +123 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=1856,device_name=NVIDIA_L40S.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H20-3e.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H20.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=352,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +122 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_GB200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H20-3e.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H20.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=704,device_name=NVIDIA_B200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=704,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +114 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=AMD_Instinct_MI308X.json +213 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_GB200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_H20.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=8960,device_name=NVIDIA_H100_80GB_HBM3,dtype=bf16.json +82 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=8960,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +82 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=928,device_name=NVIDIA_H100_80GB_HBM3.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=928,device_name=NVIDIA_L40S.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=128,N=96,device_name=NVIDIA_H20.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_B200,dtype=fp8_w8a8.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_B200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_H100.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1344,device_name=NVIDIA_A100-SXM4-40GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1344,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1792,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=2048,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=2688,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=float8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=3072,device_name=NVIDIA_H200,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=3200,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +130 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=6400,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +130 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=float8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +130 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=192,device_name=AMD_Instinct_MI300X.json +201 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=192,device_name=AMD_Instinct_MI350_OAM,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=192,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=192,device_name=NVIDIA_H20-3e.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=320,device_name=NVIDIA_H20-3e.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=384,device_name=AMD_Instinct_MI350_OAM,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=384,device_name=AMD_Instinct_MI355_OAM,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=640,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=640,device_name=NVIDIA_GB200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=160,N=640,device_name=NVIDIA_H100,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=20,N=2560,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=20,N=2560,device_name=NVIDIA_GB200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=20,N=2560,device_name=NVIDIA_H100,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=20,N=2560,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=1024,device_name=AMD_Instinct_MI325X,block_shape=[128,128].json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=1024,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=NVIDIA_H20-3e,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=256,N=64,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=32,N=1408,device_name=NVIDIA_B200.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=32,N=2048,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=32,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=384,N=128,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=384,N=128,device_name=NVIDIA_GB200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=384,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=384,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=384,N=256,device_name=NVIDIA_GB200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=40,N=1536,device_name=NVIDIA_B200,dtype=fp8_w8a8.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=40,N=2560,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=40,N=2560,device_name=NVIDIA_GB200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=40,N=2560,device_name=NVIDIA_H100,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_A100-SXM4-80GB.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_B200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_GB200,dtype=fp8_w8a8.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_H20-3e.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_B200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_GB200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_H20-3e.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_B200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_GB200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_H20-3e.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_A100-SXM4-80GB.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_B200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_H20-3e.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=60,N=1408,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=60,N=176,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=60,N=352,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=60,N=704,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=62,N=128,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=62,N=256,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=62,N=256,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=62,N=512,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=62,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=1408,device_name=NVIDIA_B200.json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=1536,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=2560,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=2560,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=2560,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=3072,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=3072,device_name=NVIDIA_H20.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=384,device_name=NVIDIA_H20.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=768,device_name=NVIDIA_H100_PCIe,dtype=fp8_w8a8,block_shape=[128,128].json +147 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=768,device_name=NVIDIA_H20.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=896,device_name=NVIDIA_H20.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=8960,device_name=NVIDIA_H100_80GB_HBM3,dtype=bf16.json +82 -0
- vllm/model_executor/layers/fused_moe/configs/E=64,N=8960,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +82 -0
- vllm/model_executor/layers/fused_moe/configs/E=72,N=192,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=72,N=384,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=72,N=384,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=72,N=768,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=72,N=768,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI325X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +138 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI325X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI325X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_A100-SXM4-40GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI325X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +154 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI325X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_L40S.json +173 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI325X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI325X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H200.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI300X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI325X.json +200 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
- vllm/model_executor/layers/fused_moe/configs/README +12 -0
- vllm/model_executor/layers/fused_moe/cpu_fused_moe.py +354 -0
- vllm/model_executor/layers/fused_moe/cutlass_moe.py +1052 -0
- vllm/model_executor/layers/fused_moe/deep_gemm_moe.py +387 -0
- vllm/model_executor/layers/fused_moe/deep_gemm_utils.py +416 -0
- vllm/model_executor/layers/fused_moe/deepep_ht_prepare_finalize.py +420 -0
- vllm/model_executor/layers/fused_moe/deepep_ll_prepare_finalize.py +367 -0
- vllm/model_executor/layers/fused_moe/flashinfer_cutlass_moe.py +307 -0
- vllm/model_executor/layers/fused_moe/flashinfer_cutlass_prepare_finalize.py +362 -0
- vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py +192 -0
- vllm/model_executor/layers/fused_moe/fused_batched_moe.py +1012 -0
- vllm/model_executor/layers/fused_moe/fused_marlin_moe.py +792 -0
- vllm/model_executor/layers/fused_moe/fused_moe.py +2175 -0
- vllm/model_executor/layers/fused_moe/fused_moe_method_base.py +112 -0
- vllm/model_executor/layers/fused_moe/fused_moe_modular_method.py +164 -0
- vllm/model_executor/layers/fused_moe/gpt_oss_triton_kernels_moe.py +316 -0
- vllm/model_executor/layers/fused_moe/layer.py +1944 -0
- vllm/model_executor/layers/fused_moe/modular_kernel.py +1222 -0
- vllm/model_executor/layers/fused_moe/moe_align_block_size.py +174 -0
- vllm/model_executor/layers/fused_moe/moe_pallas.py +83 -0
- vllm/model_executor/layers/fused_moe/moe_permute_unpermute.py +229 -0
- vllm/model_executor/layers/fused_moe/moe_torch_iterative.py +60 -0
- vllm/model_executor/layers/fused_moe/pplx_prepare_finalize.py +362 -0
- vllm/model_executor/layers/fused_moe/prepare_finalize.py +77 -0
- vllm/model_executor/layers/fused_moe/rocm_aiter_fused_moe.py +265 -0
- vllm/model_executor/layers/fused_moe/routing_simulator.py +310 -0
- vllm/model_executor/layers/fused_moe/shared_fused_moe.py +97 -0
- vllm/model_executor/layers/fused_moe/topk_weight_and_reduce.py +171 -0
- vllm/model_executor/layers/fused_moe/triton_deep_gemm_moe.py +163 -0
- vllm/model_executor/layers/fused_moe/trtllm_moe.py +143 -0
- vllm/model_executor/layers/fused_moe/unquantized_fused_moe_method.py +578 -0
- vllm/model_executor/layers/fused_moe/utils.py +332 -0
- vllm/model_executor/layers/kda.py +448 -0
- vllm/model_executor/layers/layernorm.py +442 -0
- vllm/model_executor/layers/lightning_attn.py +729 -0
- vllm/model_executor/layers/linear.py +1424 -0
- vllm/model_executor/layers/logits_processor.py +106 -0
- vllm/model_executor/layers/mamba/__init__.py +0 -0
- vllm/model_executor/layers/mamba/abstract.py +71 -0
- vllm/model_executor/layers/mamba/linear_attn.py +402 -0
- vllm/model_executor/layers/mamba/mamba_mixer.py +535 -0
- vllm/model_executor/layers/mamba/mamba_mixer2.py +928 -0
- vllm/model_executor/layers/mamba/mamba_utils.py +225 -0
- vllm/model_executor/layers/mamba/ops/__init__.py +0 -0
- vllm/model_executor/layers/mamba/ops/causal_conv1d.py +1240 -0
- vllm/model_executor/layers/mamba/ops/layernorm_gated.py +172 -0
- vllm/model_executor/layers/mamba/ops/mamba_ssm.py +478 -0
- vllm/model_executor/layers/mamba/ops/ssd_bmm.py +211 -0
- vllm/model_executor/layers/mamba/ops/ssd_chunk_scan.py +456 -0
- vllm/model_executor/layers/mamba/ops/ssd_chunk_state.py +700 -0
- vllm/model_executor/layers/mamba/ops/ssd_combined.py +230 -0
- vllm/model_executor/layers/mamba/ops/ssd_state_passing.py +157 -0
- vllm/model_executor/layers/mamba/short_conv.py +264 -0
- vllm/model_executor/layers/mla.py +168 -0
- vllm/model_executor/layers/pooler.py +817 -0
- vllm/model_executor/layers/quantization/__init__.py +174 -0
- vllm/model_executor/layers/quantization/auto_round.py +454 -0
- vllm/model_executor/layers/quantization/awq.py +277 -0
- vllm/model_executor/layers/quantization/awq_marlin.py +659 -0
- vllm/model_executor/layers/quantization/awq_triton.py +337 -0
- vllm/model_executor/layers/quantization/base_config.py +170 -0
- vllm/model_executor/layers/quantization/bitblas.py +502 -0
- vllm/model_executor/layers/quantization/bitsandbytes.py +658 -0
- vllm/model_executor/layers/quantization/compressed_tensors/__init__.py +3 -0
- vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py +914 -0
- vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py +2284 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/__init__.py +35 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_24.py +392 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_scheme.py +55 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a16_24.py +176 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a16_nvfp4.py +124 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_nvfp4.py +218 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a8_fp8.py +183 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a8_int.py +153 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a16_fp8.py +138 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py +200 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py +125 -0
- vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16.py +219 -0
- vllm/model_executor/layers/quantization/compressed_tensors/transform/__init__.py +0 -0
- vllm/model_executor/layers/quantization/compressed_tensors/transform/linear.py +260 -0
- vllm/model_executor/layers/quantization/compressed_tensors/transform/module.py +173 -0
- vllm/model_executor/layers/quantization/compressed_tensors/transform/schemes/__init__.py +0 -0
- vllm/model_executor/layers/quantization/compressed_tensors/transform/schemes/linear_qutlass_nvfp4.py +64 -0
- vllm/model_executor/layers/quantization/compressed_tensors/transform/utils.py +13 -0
- vllm/model_executor/layers/quantization/compressed_tensors/triton_scaled_mm.py +224 -0
- vllm/model_executor/layers/quantization/compressed_tensors/utils.py +216 -0
- vllm/model_executor/layers/quantization/deepspeedfp.py +218 -0
- vllm/model_executor/layers/quantization/experts_int8.py +240 -0
- vllm/model_executor/layers/quantization/fbgemm_fp8.py +195 -0
- vllm/model_executor/layers/quantization/fp8.py +1333 -0
- vllm/model_executor/layers/quantization/fp_quant.py +420 -0
- vllm/model_executor/layers/quantization/gguf.py +643 -0
- vllm/model_executor/layers/quantization/gptq.py +393 -0
- vllm/model_executor/layers/quantization/gptq_bitblas.py +482 -0
- vllm/model_executor/layers/quantization/gptq_marlin.py +789 -0
- vllm/model_executor/layers/quantization/gptq_marlin_24.py +320 -0
- vllm/model_executor/layers/quantization/hqq_marlin.py +371 -0
- vllm/model_executor/layers/quantization/inc.py +65 -0
- vllm/model_executor/layers/quantization/input_quant_fp8.py +171 -0
- vllm/model_executor/layers/quantization/ipex_quant.py +467 -0
- vllm/model_executor/layers/quantization/kernels/__init__.py +0 -0
- vllm/model_executor/layers/quantization/kernels/mixed_precision/MPLinearKernel.py +94 -0
- vllm/model_executor/layers/quantization/kernels/mixed_precision/__init__.py +105 -0
- vllm/model_executor/layers/quantization/kernels/mixed_precision/allspark.py +115 -0
- vllm/model_executor/layers/quantization/kernels/mixed_precision/bitblas.py +323 -0
- vllm/model_executor/layers/quantization/kernels/mixed_precision/conch.py +98 -0
- vllm/model_executor/layers/quantization/kernels/mixed_precision/cutlass.py +119 -0
- vllm/model_executor/layers/quantization/kernels/mixed_precision/dynamic_4bit.py +111 -0
- vllm/model_executor/layers/quantization/kernels/mixed_precision/exllama.py +161 -0
- vllm/model_executor/layers/quantization/kernels/mixed_precision/machete.py +159 -0
- vllm/model_executor/layers/quantization/kernels/mixed_precision/marlin.py +166 -0
- vllm/model_executor/layers/quantization/kernels/scaled_mm/ScaledMMLinearKernel.py +73 -0
- vllm/model_executor/layers/quantization/kernels/scaled_mm/__init__.py +97 -0
- vllm/model_executor/layers/quantization/kernels/scaled_mm/aiter.py +120 -0
- vllm/model_executor/layers/quantization/kernels/scaled_mm/cpu.py +219 -0
- vllm/model_executor/layers/quantization/kernels/scaled_mm/cutlass.py +140 -0
- vllm/model_executor/layers/quantization/kernels/scaled_mm/triton.py +42 -0
- vllm/model_executor/layers/quantization/kernels/scaled_mm/xla.py +105 -0
- vllm/model_executor/layers/quantization/kv_cache.py +146 -0
- vllm/model_executor/layers/quantization/modelopt.py +1788 -0
- vllm/model_executor/layers/quantization/moe_wna16.py +541 -0
- vllm/model_executor/layers/quantization/mxfp4.py +1162 -0
- vllm/model_executor/layers/quantization/petit.py +320 -0
- vllm/model_executor/layers/quantization/ptpc_fp8.py +137 -0
- vllm/model_executor/layers/quantization/quark/__init__.py +0 -0
- vllm/model_executor/layers/quantization/quark/quark.py +528 -0
- vllm/model_executor/layers/quantization/quark/quark_moe.py +683 -0
- vllm/model_executor/layers/quantization/quark/schemes/__init__.py +9 -0
- vllm/model_executor/layers/quantization/quark/schemes/quark_ocp_mx.py +306 -0
- vllm/model_executor/layers/quantization/quark/schemes/quark_scheme.py +55 -0
- vllm/model_executor/layers/quantization/quark/schemes/quark_w8a8_fp8.py +179 -0
- vllm/model_executor/layers/quantization/quark/schemes/quark_w8a8_int8.py +139 -0
- vllm/model_executor/layers/quantization/quark/utils.py +105 -0
- vllm/model_executor/layers/quantization/qutlass_utils.py +185 -0
- vllm/model_executor/layers/quantization/rtn.py +652 -0
- vllm/model_executor/layers/quantization/schema.py +90 -0
- vllm/model_executor/layers/quantization/torchao.py +380 -0
- vllm/model_executor/layers/quantization/tpu_int8.py +139 -0
- vllm/model_executor/layers/quantization/utils/__init__.py +6 -0
- vllm/model_executor/layers/quantization/utils/allspark_utils.py +67 -0
- vllm/model_executor/layers/quantization/utils/bitblas_utils.py +229 -0
- vllm/model_executor/layers/quantization/utils/configs/N=12288,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=12288,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2112,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2112,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=1536,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4096,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4096,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +18 -0
- vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json +146 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=NVIDIA_L20,dtype=fp8_w8a8,block_shape=[128,128].json +26 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=8192,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=7168,K=8192,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=8192,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=8192,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/N=8192,K=1536,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json +164 -0
- vllm/model_executor/layers/quantization/utils/configs/README.md +3 -0
- vllm/model_executor/layers/quantization/utils/flashinfer_fp4_moe.py +89 -0
- vllm/model_executor/layers/quantization/utils/flashinfer_utils.py +298 -0
- vllm/model_executor/layers/quantization/utils/fp8_utils.py +1203 -0
- vllm/model_executor/layers/quantization/utils/gptq_utils.py +158 -0
- vllm/model_executor/layers/quantization/utils/int8_utils.py +489 -0
- vllm/model_executor/layers/quantization/utils/layer_utils.py +41 -0
- vllm/model_executor/layers/quantization/utils/machete_utils.py +56 -0
- vllm/model_executor/layers/quantization/utils/marlin_utils.py +575 -0
- vllm/model_executor/layers/quantization/utils/marlin_utils_fp4.py +397 -0
- vllm/model_executor/layers/quantization/utils/marlin_utils_fp8.py +351 -0
- vllm/model_executor/layers/quantization/utils/marlin_utils_test.py +161 -0
- vllm/model_executor/layers/quantization/utils/marlin_utils_test_24.py +467 -0
- vllm/model_executor/layers/quantization/utils/mxfp4_utils.py +181 -0
- vllm/model_executor/layers/quantization/utils/mxfp6_utils.py +142 -0
- vllm/model_executor/layers/quantization/utils/mxfp8_utils.py +24 -0
- vllm/model_executor/layers/quantization/utils/nvfp4_emulation_utils.py +142 -0
- vllm/model_executor/layers/quantization/utils/nvfp4_moe_support.py +63 -0
- vllm/model_executor/layers/quantization/utils/ocp_mx_utils.py +51 -0
- vllm/model_executor/layers/quantization/utils/petit_utils.py +124 -0
- vllm/model_executor/layers/quantization/utils/quant_utils.py +687 -0
- vllm/model_executor/layers/quantization/utils/w8a8_utils.py +516 -0
- vllm/model_executor/layers/resampler.py +283 -0
- vllm/model_executor/layers/rotary_embedding/__init__.py +278 -0
- vllm/model_executor/layers/rotary_embedding/base.py +235 -0
- vllm/model_executor/layers/rotary_embedding/common.py +188 -0
- vllm/model_executor/layers/rotary_embedding/deepseek_scaling_rope.py +165 -0
- vllm/model_executor/layers/rotary_embedding/dual_chunk_rope.py +215 -0
- vllm/model_executor/layers/rotary_embedding/dynamic_ntk_alpha_rope.py +43 -0
- vllm/model_executor/layers/rotary_embedding/dynamic_ntk_scaling_rope.py +68 -0
- vllm/model_executor/layers/rotary_embedding/ernie45_vl_rope.py +75 -0
- vllm/model_executor/layers/rotary_embedding/linear_scaling_rope.py +115 -0
- vllm/model_executor/layers/rotary_embedding/llama3_rope.py +54 -0
- vllm/model_executor/layers/rotary_embedding/llama4_vision_rope.py +80 -0
- vllm/model_executor/layers/rotary_embedding/mrope.py +397 -0
- vllm/model_executor/layers/rotary_embedding/ntk_scaling_rope.py +47 -0
- vllm/model_executor/layers/rotary_embedding/phi3_long_rope_scaled_rope.py +159 -0
- vllm/model_executor/layers/rotary_embedding/yarn_scaling_rope.py +81 -0
- vllm/model_executor/layers/utils.py +251 -0
- vllm/model_executor/layers/vocab_parallel_embedding.py +558 -0
- vllm/model_executor/model_loader/__init__.py +148 -0
- vllm/model_executor/model_loader/base_loader.py +57 -0
- vllm/model_executor/model_loader/bitsandbytes_loader.py +822 -0
- vllm/model_executor/model_loader/default_loader.py +327 -0
- vllm/model_executor/model_loader/dummy_loader.py +28 -0
- vllm/model_executor/model_loader/gguf_loader.py +176 -0
- vllm/model_executor/model_loader/online_quantization.py +224 -0
- vllm/model_executor/model_loader/runai_streamer_loader.py +116 -0
- vllm/model_executor/model_loader/sharded_state_loader.py +206 -0
- vllm/model_executor/model_loader/tensorizer.py +790 -0
- vllm/model_executor/model_loader/tensorizer_loader.py +151 -0
- vllm/model_executor/model_loader/tpu.py +118 -0
- vllm/model_executor/model_loader/utils.py +288 -0
- vllm/model_executor/model_loader/weight_utils.py +1084 -0
- vllm/model_executor/models/__init__.py +44 -0
- vllm/model_executor/models/adapters.py +543 -0
- vllm/model_executor/models/afmoe.py +711 -0
- vllm/model_executor/models/aimv2.py +247 -0
- vllm/model_executor/models/apertus.py +587 -0
- vllm/model_executor/models/arcee.py +439 -0
- vllm/model_executor/models/arctic.py +635 -0
- vllm/model_executor/models/aria.py +655 -0
- vllm/model_executor/models/aya_vision.py +450 -0
- vllm/model_executor/models/baichuan.py +496 -0
- vllm/model_executor/models/bailing_moe.py +646 -0
- vllm/model_executor/models/bamba.py +522 -0
- vllm/model_executor/models/bee.py +157 -0
- vllm/model_executor/models/bert.py +925 -0
- vllm/model_executor/models/bert_with_rope.py +732 -0
- vllm/model_executor/models/blip.py +349 -0
- vllm/model_executor/models/blip2.py +695 -0
- vllm/model_executor/models/bloom.py +390 -0
- vllm/model_executor/models/chameleon.py +1120 -0
- vllm/model_executor/models/chatglm.py +498 -0
- vllm/model_executor/models/clip.py +965 -0
- vllm/model_executor/models/cohere2_vision.py +472 -0
- vllm/model_executor/models/commandr.py +473 -0
- vllm/model_executor/models/config.py +503 -0
- vllm/model_executor/models/dbrx.py +482 -0
- vllm/model_executor/models/deepencoder.py +673 -0
- vllm/model_executor/models/deepseek_eagle.py +260 -0
- vllm/model_executor/models/deepseek_mtp.py +360 -0
- vllm/model_executor/models/deepseek_ocr.py +593 -0
- vllm/model_executor/models/deepseek_v2.py +1649 -0
- vllm/model_executor/models/deepseek_vl2.py +655 -0
- vllm/model_executor/models/dots1.py +574 -0
- vllm/model_executor/models/dots_ocr.py +900 -0
- vllm/model_executor/models/ernie45.py +53 -0
- vllm/model_executor/models/ernie45_moe.py +759 -0
- vllm/model_executor/models/ernie45_vl.py +1742 -0
- vllm/model_executor/models/ernie45_vl_moe.py +803 -0
- vllm/model_executor/models/ernie_mtp.py +279 -0
- vllm/model_executor/models/exaone.py +545 -0
- vllm/model_executor/models/exaone4.py +531 -0
- vllm/model_executor/models/fairseq2_llama.py +154 -0
- vllm/model_executor/models/falcon.py +545 -0
- vllm/model_executor/models/falcon_h1.py +685 -0
- vllm/model_executor/models/flex_olmo.py +155 -0
- vllm/model_executor/models/fuyu.py +373 -0
- vllm/model_executor/models/gemma.py +426 -0
- vllm/model_executor/models/gemma2.py +439 -0
- vllm/model_executor/models/gemma3.py +571 -0
- vllm/model_executor/models/gemma3_mm.py +741 -0
- vllm/model_executor/models/gemma3n.py +1165 -0
- vllm/model_executor/models/gemma3n_mm.py +811 -0
- vllm/model_executor/models/glm.py +23 -0
- vllm/model_executor/models/glm4.py +305 -0
- vllm/model_executor/models/glm4_1v.py +1821 -0
- vllm/model_executor/models/glm4_moe.py +747 -0
- vllm/model_executor/models/glm4_moe_mtp.py +359 -0
- vllm/model_executor/models/glm4v.py +784 -0
- vllm/model_executor/models/gpt2.py +397 -0
- vllm/model_executor/models/gpt_bigcode.py +339 -0
- vllm/model_executor/models/gpt_j.py +346 -0
- vllm/model_executor/models/gpt_neox.py +344 -0
- vllm/model_executor/models/gpt_oss.py +738 -0
- vllm/model_executor/models/granite.py +516 -0
- vllm/model_executor/models/granite_speech.py +913 -0
- vllm/model_executor/models/granitemoe.py +569 -0
- vllm/model_executor/models/granitemoehybrid.py +709 -0
- vllm/model_executor/models/granitemoeshared.py +333 -0
- vllm/model_executor/models/gritlm.py +245 -0
- vllm/model_executor/models/grok1.py +558 -0
- vllm/model_executor/models/h2ovl.py +554 -0
- vllm/model_executor/models/hunyuan_v1.py +1053 -0
- vllm/model_executor/models/hyperclovax_vision.py +1166 -0
- vllm/model_executor/models/idefics2_vision_model.py +426 -0
- vllm/model_executor/models/idefics3.py +717 -0
- vllm/model_executor/models/interfaces.py +1092 -0
- vllm/model_executor/models/interfaces_base.py +214 -0
- vllm/model_executor/models/intern_vit.py +453 -0
- vllm/model_executor/models/internlm2.py +460 -0
- vllm/model_executor/models/internlm2_ve.py +142 -0
- vllm/model_executor/models/interns1.py +830 -0
- vllm/model_executor/models/interns1_vit.py +432 -0
- vllm/model_executor/models/internvl.py +1452 -0
- vllm/model_executor/models/jais.py +397 -0
- vllm/model_executor/models/jamba.py +610 -0
- vllm/model_executor/models/jina_vl.py +147 -0
- vllm/model_executor/models/keye.py +1761 -0
- vllm/model_executor/models/keye_vl1_5.py +726 -0
- vllm/model_executor/models/kimi_linear.py +663 -0
- vllm/model_executor/models/kimi_vl.py +578 -0
- vllm/model_executor/models/lfm2.py +532 -0
- vllm/model_executor/models/lfm2_moe.py +762 -0
- vllm/model_executor/models/lightonocr.py +195 -0
- vllm/model_executor/models/llama.py +732 -0
- vllm/model_executor/models/llama4.py +859 -0
- vllm/model_executor/models/llama4_eagle.py +223 -0
- vllm/model_executor/models/llama_eagle.py +218 -0
- vllm/model_executor/models/llama_eagle3.py +367 -0
- vllm/model_executor/models/llava.py +842 -0
- vllm/model_executor/models/llava_next.py +583 -0
- vllm/model_executor/models/llava_next_video.py +467 -0
- vllm/model_executor/models/llava_onevision.py +923 -0
- vllm/model_executor/models/longcat_flash.py +749 -0
- vllm/model_executor/models/longcat_flash_mtp.py +349 -0
- vllm/model_executor/models/mamba.py +276 -0
- vllm/model_executor/models/mamba2.py +289 -0
- vllm/model_executor/models/medusa.py +179 -0
- vllm/model_executor/models/midashenglm.py +827 -0
- vllm/model_executor/models/mimo.py +188 -0
- vllm/model_executor/models/mimo_mtp.py +294 -0
- vllm/model_executor/models/minicpm.py +664 -0
- vllm/model_executor/models/minicpm3.py +242 -0
- vllm/model_executor/models/minicpm_eagle.py +389 -0
- vllm/model_executor/models/minicpmo.py +768 -0
- vllm/model_executor/models/minicpmv.py +1745 -0
- vllm/model_executor/models/minimax_m2.py +552 -0
- vllm/model_executor/models/minimax_text_01.py +1012 -0
- vllm/model_executor/models/minimax_vl_01.py +396 -0
- vllm/model_executor/models/mistral3.py +637 -0
- vllm/model_executor/models/mixtral.py +621 -0
- vllm/model_executor/models/mllama4.py +1147 -0
- vllm/model_executor/models/mlp_speculator.py +235 -0
- vllm/model_executor/models/modernbert.py +450 -0
- vllm/model_executor/models/module_mapping.py +74 -0
- vllm/model_executor/models/molmo.py +1555 -0
- vllm/model_executor/models/moonvit.py +677 -0
- vllm/model_executor/models/mpt.py +335 -0
- vllm/model_executor/models/nano_nemotron_vl.py +1740 -0
- vllm/model_executor/models/nemotron.py +518 -0
- vllm/model_executor/models/nemotron_h.py +852 -0
- vllm/model_executor/models/nemotron_nas.py +491 -0
- vllm/model_executor/models/nemotron_vl.py +653 -0
- vllm/model_executor/models/nvlm_d.py +216 -0
- vllm/model_executor/models/olmo.py +414 -0
- vllm/model_executor/models/olmo2.py +454 -0
- vllm/model_executor/models/olmoe.py +498 -0
- vllm/model_executor/models/openpangu.py +1062 -0
- vllm/model_executor/models/openpangu_mtp.py +265 -0
- vllm/model_executor/models/opt.py +426 -0
- vllm/model_executor/models/orion.py +372 -0
- vllm/model_executor/models/ouro.py +516 -0
- vllm/model_executor/models/ovis.py +559 -0
- vllm/model_executor/models/ovis2_5.py +673 -0
- vllm/model_executor/models/paddleocr_vl.py +1407 -0
- vllm/model_executor/models/paligemma.py +412 -0
- vllm/model_executor/models/persimmon.py +377 -0
- vllm/model_executor/models/phi.py +374 -0
- vllm/model_executor/models/phi3.py +18 -0
- vllm/model_executor/models/phi3v.py +737 -0
- vllm/model_executor/models/phi4_multimodal.py +1447 -0
- vllm/model_executor/models/phi4mm.py +1253 -0
- vllm/model_executor/models/phi4mm_audio.py +1296 -0
- vllm/model_executor/models/phi4mm_utils.py +1907 -0
- vllm/model_executor/models/phimoe.py +675 -0
- vllm/model_executor/models/pixtral.py +1352 -0
- vllm/model_executor/models/plamo2.py +981 -0
- vllm/model_executor/models/qwen.py +368 -0
- vllm/model_executor/models/qwen2.py +541 -0
- vllm/model_executor/models/qwen2_5_omni_thinker.py +1246 -0
- vllm/model_executor/models/qwen2_5_vl.py +1613 -0
- vllm/model_executor/models/qwen2_audio.py +473 -0
- vllm/model_executor/models/qwen2_moe.py +596 -0
- vllm/model_executor/models/qwen2_rm.py +123 -0
- vllm/model_executor/models/qwen2_vl.py +1670 -0
- vllm/model_executor/models/qwen3.py +336 -0
- vllm/model_executor/models/qwen3_moe.py +744 -0
- vllm/model_executor/models/qwen3_next.py +1395 -0
- vllm/model_executor/models/qwen3_next_mtp.py +296 -0
- vllm/model_executor/models/qwen3_omni_moe_thinker.py +1721 -0
- vllm/model_executor/models/qwen3_vl.py +1673 -0
- vllm/model_executor/models/qwen3_vl_moe.py +415 -0
- vllm/model_executor/models/qwen_vl.py +802 -0
- vllm/model_executor/models/radio.py +555 -0
- vllm/model_executor/models/registry.py +1155 -0
- vllm/model_executor/models/roberta.py +259 -0
- vllm/model_executor/models/rvl.py +107 -0
- vllm/model_executor/models/seed_oss.py +497 -0
- vllm/model_executor/models/siglip.py +1174 -0
- vllm/model_executor/models/siglip2navit.py +724 -0
- vllm/model_executor/models/skyworkr1v.py +953 -0
- vllm/model_executor/models/smolvlm.py +38 -0
- vllm/model_executor/models/solar.py +502 -0
- vllm/model_executor/models/stablelm.py +359 -0
- vllm/model_executor/models/starcoder2.py +367 -0
- vllm/model_executor/models/step3_text.py +559 -0
- vllm/model_executor/models/step3_vl.py +1148 -0
- vllm/model_executor/models/swin.py +514 -0
- vllm/model_executor/models/tarsier.py +619 -0
- vllm/model_executor/models/telechat2.py +153 -0
- vllm/model_executor/models/teleflm.py +78 -0
- vllm/model_executor/models/terratorch.py +319 -0
- vllm/model_executor/models/transformers/__init__.py +127 -0
- vllm/model_executor/models/transformers/base.py +464 -0
- vllm/model_executor/models/transformers/causal.py +65 -0
- vllm/model_executor/models/transformers/legacy.py +90 -0
- vllm/model_executor/models/transformers/moe.py +318 -0
- vllm/model_executor/models/transformers/multimodal.py +411 -0
- vllm/model_executor/models/transformers/pooling.py +119 -0
- vllm/model_executor/models/transformers/utils.py +207 -0
- vllm/model_executor/models/ultravox.py +681 -0
- vllm/model_executor/models/utils.py +877 -0
- vllm/model_executor/models/vision.py +552 -0
- vllm/model_executor/models/voxtral.py +845 -0
- vllm/model_executor/models/whisper.py +959 -0
- vllm/model_executor/models/zamba2.py +986 -0
- vllm/model_executor/parameter.py +642 -0
- vllm/model_executor/utils.py +94 -0
- vllm/model_executor/warmup/__init__.py +0 -0
- vllm/model_executor/warmup/deep_gemm_warmup.py +314 -0
- vllm/model_executor/warmup/kernel_warmup.py +98 -0
- vllm/multimodal/__init__.py +40 -0
- vllm/multimodal/audio.py +118 -0
- vllm/multimodal/base.py +26 -0
- vllm/multimodal/cache.py +755 -0
- vllm/multimodal/evs.py +294 -0
- vllm/multimodal/hasher.py +106 -0
- vllm/multimodal/image.py +130 -0
- vllm/multimodal/inputs.py +1036 -0
- vllm/multimodal/parse.py +544 -0
- vllm/multimodal/processing.py +2186 -0
- vllm/multimodal/profiling.py +369 -0
- vllm/multimodal/registry.py +360 -0
- vllm/multimodal/utils.py +512 -0
- vllm/multimodal/video.py +306 -0
- vllm/outputs.py +345 -0
- vllm/platforms/__init__.py +277 -0
- vllm/platforms/cpu.py +414 -0
- vllm/platforms/cuda.py +657 -0
- vllm/platforms/interface.py +639 -0
- vllm/platforms/rocm.py +466 -0
- vllm/platforms/tpu.py +276 -0
- vllm/platforms/xpu.py +274 -0
- vllm/plugins/__init__.py +78 -0
- vllm/plugins/io_processors/__init__.py +68 -0
- vllm/plugins/io_processors/interface.py +77 -0
- vllm/plugins/lora_resolvers/__init__.py +0 -0
- vllm/plugins/lora_resolvers/filesystem_resolver.py +52 -0
- vllm/pooling_params.py +228 -0
- vllm/profiler/__init__.py +0 -0
- vllm/profiler/gpu_profiler.py +37 -0
- vllm/profiler/layerwise_profile.py +392 -0
- vllm/profiler/utils.py +151 -0
- vllm/py.typed +2 -0
- vllm/ray/__init__.py +0 -0
- vllm/ray/lazy_utils.py +26 -0
- vllm/ray/ray_env.py +79 -0
- vllm/reasoning/__init__.py +92 -0
- vllm/reasoning/abs_reasoning_parsers.py +290 -0
- vllm/reasoning/basic_parsers.py +162 -0
- vllm/reasoning/deepseek_r1_reasoning_parser.py +67 -0
- vllm/reasoning/deepseek_v3_reasoning_parser.py +62 -0
- vllm/reasoning/ernie45_reasoning_parser.py +165 -0
- vllm/reasoning/glm4_moe_reasoning_parser.py +171 -0
- vllm/reasoning/gptoss_reasoning_parser.py +173 -0
- vllm/reasoning/granite_reasoning_parser.py +363 -0
- vllm/reasoning/hunyuan_a13b_reasoning_parser.py +237 -0
- vllm/reasoning/identity_reasoning_parser.py +58 -0
- vllm/reasoning/minimax_m2_reasoning_parser.py +67 -0
- vllm/reasoning/mistral_reasoning_parser.py +55 -0
- vllm/reasoning/olmo3_reasoning_parser.py +302 -0
- vllm/reasoning/qwen3_reasoning_parser.py +67 -0
- vllm/reasoning/seedoss_reasoning_parser.py +27 -0
- vllm/reasoning/step3_reasoning_parser.py +107 -0
- vllm/sampling_params.py +669 -0
- vllm/scalar_type.py +355 -0
- vllm/scripts.py +17 -0
- vllm/sequence.py +98 -0
- vllm/tasks.py +13 -0
- vllm/third_party/__init__.py +0 -0
- vllm/third_party/pynvml.py +6140 -0
- vllm/tracing.py +135 -0
- vllm/transformers_utils/__init__.py +26 -0
- vllm/transformers_utils/chat_templates/__init__.py +5 -0
- vllm/transformers_utils/chat_templates/registry.py +73 -0
- vllm/transformers_utils/chat_templates/template_basic.jinja +3 -0
- vllm/transformers_utils/chat_templates/template_blip2.jinja +11 -0
- vllm/transformers_utils/chat_templates/template_chatml.jinja +10 -0
- vllm/transformers_utils/chat_templates/template_deepseek_ocr.jinja +14 -0
- vllm/transformers_utils/chat_templates/template_deepseek_vl2.jinja +23 -0
- vllm/transformers_utils/chat_templates/template_fuyu.jinja +3 -0
- vllm/transformers_utils/chat_templates/template_minicpmv45.jinja +93 -0
- vllm/transformers_utils/config.py +1203 -0
- vllm/transformers_utils/config_parser_base.py +20 -0
- vllm/transformers_utils/configs/__init__.py +70 -0
- vllm/transformers_utils/configs/afmoe.py +84 -0
- vllm/transformers_utils/configs/arctic.py +206 -0
- vllm/transformers_utils/configs/chatglm.py +75 -0
- vllm/transformers_utils/configs/deepseek_vl2.py +126 -0
- vllm/transformers_utils/configs/dotsocr.py +71 -0
- vllm/transformers_utils/configs/eagle.py +84 -0
- vllm/transformers_utils/configs/falcon.py +89 -0
- vllm/transformers_utils/configs/flex_olmo.py +77 -0
- vllm/transformers_utils/configs/jais.py +243 -0
- vllm/transformers_utils/configs/kimi_linear.py +144 -0
- vllm/transformers_utils/configs/kimi_vl.py +38 -0
- vllm/transformers_utils/configs/lfm2_moe.py +159 -0
- vllm/transformers_utils/configs/medusa.py +65 -0
- vllm/transformers_utils/configs/midashenglm.py +103 -0
- vllm/transformers_utils/configs/mistral.py +174 -0
- vllm/transformers_utils/configs/mlp_speculator.py +69 -0
- vllm/transformers_utils/configs/moonvit.py +33 -0
- vllm/transformers_utils/configs/nemotron.py +212 -0
- vllm/transformers_utils/configs/nemotron_h.py +282 -0
- vllm/transformers_utils/configs/olmo3.py +79 -0
- vllm/transformers_utils/configs/ovis.py +182 -0
- vllm/transformers_utils/configs/qwen3_next.py +274 -0
- vllm/transformers_utils/configs/radio.py +89 -0
- vllm/transformers_utils/configs/speculators/__init__.py +2 -0
- vllm/transformers_utils/configs/speculators/algos.py +38 -0
- vllm/transformers_utils/configs/speculators/base.py +114 -0
- vllm/transformers_utils/configs/step3_vl.py +174 -0
- vllm/transformers_utils/configs/ultravox.py +118 -0
- vllm/transformers_utils/detokenizer_utils.py +198 -0
- vllm/transformers_utils/dynamic_module.py +59 -0
- vllm/transformers_utils/processor.py +402 -0
- vllm/transformers_utils/processors/__init__.py +15 -0
- vllm/transformers_utils/processors/deepseek_ocr.py +438 -0
- vllm/transformers_utils/processors/deepseek_vl2.py +406 -0
- vllm/transformers_utils/processors/ovis.py +453 -0
- vllm/transformers_utils/processors/ovis2_5.py +468 -0
- vllm/transformers_utils/runai_utils.py +104 -0
- vllm/transformers_utils/s3_utils.py +95 -0
- vllm/transformers_utils/tokenizer.py +293 -0
- vllm/transformers_utils/tokenizer_base.py +155 -0
- vllm/transformers_utils/tokenizers/__init__.py +16 -0
- vllm/transformers_utils/tokenizers/mistral.py +502 -0
- vllm/transformers_utils/utils.py +130 -0
- vllm/triton_utils/__init__.py +19 -0
- vllm/triton_utils/importing.py +103 -0
- vllm/usage/__init__.py +0 -0
- vllm/usage/usage_lib.py +294 -0
- vllm/utils/__init__.py +82 -0
- vllm/utils/argparse_utils.py +487 -0
- vllm/utils/async_utils.py +303 -0
- vllm/utils/cache.py +214 -0
- vllm/utils/collection_utils.py +139 -0
- vllm/utils/counter.py +45 -0
- vllm/utils/deep_gemm.py +391 -0
- vllm/utils/flashinfer.py +490 -0
- vllm/utils/func_utils.py +236 -0
- vllm/utils/gc_utils.py +147 -0
- vllm/utils/hashing.py +63 -0
- vllm/utils/import_utils.py +411 -0
- vllm/utils/jsontree.py +165 -0
- vllm/utils/math_utils.py +32 -0
- vllm/utils/mem_constants.py +13 -0
- vllm/utils/mem_utils.py +232 -0
- vllm/utils/nccl.py +64 -0
- vllm/utils/network_utils.py +331 -0
- vllm/utils/platform_utils.py +59 -0
- vllm/utils/profiling.py +56 -0
- vllm/utils/registry.py +49 -0
- vllm/utils/serial_utils.py +169 -0
- vllm/utils/system_utils.py +229 -0
- vllm/utils/tensor_schema.py +255 -0
- vllm/utils/torch_utils.py +657 -0
- vllm/v1/__init__.py +0 -0
- vllm/v1/attention/__init__.py +0 -0
- vllm/v1/attention/backends/__init__.py +0 -0
- vllm/v1/attention/backends/cpu_attn.py +496 -0
- vllm/v1/attention/backends/flash_attn.py +1028 -0
- vllm/v1/attention/backends/flashinfer.py +1572 -0
- vllm/v1/attention/backends/flex_attention.py +926 -0
- vllm/v1/attention/backends/gdn_attn.py +387 -0
- vllm/v1/attention/backends/linear_attn.py +74 -0
- vllm/v1/attention/backends/mamba1_attn.py +165 -0
- vllm/v1/attention/backends/mamba2_attn.py +354 -0
- vllm/v1/attention/backends/mamba_attn.py +115 -0
- vllm/v1/attention/backends/mla/__init__.py +0 -0
- vllm/v1/attention/backends/mla/common.py +2031 -0
- vllm/v1/attention/backends/mla/cutlass_mla.py +275 -0
- vllm/v1/attention/backends/mla/flashattn_mla.py +337 -0
- vllm/v1/attention/backends/mla/flashinfer_mla.py +171 -0
- vllm/v1/attention/backends/mla/flashmla.py +314 -0
- vllm/v1/attention/backends/mla/flashmla_sparse.py +548 -0
- vllm/v1/attention/backends/mla/indexer.py +362 -0
- vllm/v1/attention/backends/mla/rocm_aiter_mla.py +294 -0
- vllm/v1/attention/backends/mla/triton_mla.py +171 -0
- vllm/v1/attention/backends/pallas.py +436 -0
- vllm/v1/attention/backends/rocm_aiter_fa.py +816 -0
- vllm/v1/attention/backends/rocm_aiter_unified_attn.py +196 -0
- vllm/v1/attention/backends/rocm_attn.py +362 -0
- vllm/v1/attention/backends/short_conv_attn.py +105 -0
- vllm/v1/attention/backends/tree_attn.py +425 -0
- vllm/v1/attention/backends/triton_attn.py +373 -0
- vllm/v1/attention/backends/utils.py +1116 -0
- vllm/v1/attention/backends/xformers.py +417 -0
- vllm/v1/core/__init__.py +0 -0
- vllm/v1/core/block_pool.py +428 -0
- vllm/v1/core/encoder_cache_manager.py +343 -0
- vllm/v1/core/kv_cache_coordinator.py +480 -0
- vllm/v1/core/kv_cache_manager.py +420 -0
- vllm/v1/core/kv_cache_utils.py +1340 -0
- vllm/v1/core/sched/__init__.py +0 -0
- vllm/v1/core/sched/async_scheduler.py +62 -0
- vllm/v1/core/sched/interface.py +181 -0
- vllm/v1/core/sched/output.py +202 -0
- vllm/v1/core/sched/request_queue.py +221 -0
- vllm/v1/core/sched/scheduler.py +1617 -0
- vllm/v1/core/sched/utils.py +72 -0
- vllm/v1/core/single_type_kv_cache_manager.py +736 -0
- vllm/v1/cudagraph_dispatcher.py +148 -0
- vllm/v1/engine/__init__.py +206 -0
- vllm/v1/engine/async_llm.py +797 -0
- vllm/v1/engine/coordinator.py +377 -0
- vllm/v1/engine/core.py +1420 -0
- vllm/v1/engine/core_client.py +1400 -0
- vllm/v1/engine/detokenizer.py +351 -0
- vllm/v1/engine/exceptions.py +18 -0
- vllm/v1/engine/llm_engine.py +408 -0
- vllm/v1/engine/logprobs.py +182 -0
- vllm/v1/engine/output_processor.py +642 -0
- vllm/v1/engine/parallel_sampling.py +145 -0
- vllm/v1/engine/processor.py +621 -0
- vllm/v1/engine/utils.py +1072 -0
- vllm/v1/executor/__init__.py +6 -0
- vllm/v1/executor/abstract.py +352 -0
- vllm/v1/executor/multiproc_executor.py +877 -0
- vllm/v1/executor/ray_distributed_executor.py +8 -0
- vllm/v1/executor/ray_executor.py +626 -0
- vllm/v1/executor/ray_utils.py +465 -0
- vllm/v1/executor/uniproc_executor.py +183 -0
- vllm/v1/kv_cache_interface.py +403 -0
- vllm/v1/kv_offload/__init__.py +0 -0
- vllm/v1/kv_offload/abstract.py +161 -0
- vllm/v1/kv_offload/arc_manager.py +237 -0
- vllm/v1/kv_offload/backend.py +97 -0
- vllm/v1/kv_offload/backends/__init__.py +0 -0
- vllm/v1/kv_offload/backends/cpu.py +62 -0
- vllm/v1/kv_offload/cpu.py +93 -0
- vllm/v1/kv_offload/factory.py +56 -0
- vllm/v1/kv_offload/lru_manager.py +139 -0
- vllm/v1/kv_offload/mediums.py +39 -0
- vllm/v1/kv_offload/spec.py +62 -0
- vllm/v1/kv_offload/worker/__init__.py +0 -0
- vllm/v1/kv_offload/worker/cpu_gpu.py +185 -0
- vllm/v1/kv_offload/worker/worker.py +144 -0
- vllm/v1/metrics/__init__.py +0 -0
- vllm/v1/metrics/loggers.py +1238 -0
- vllm/v1/metrics/prometheus.py +82 -0
- vllm/v1/metrics/ray_wrappers.py +169 -0
- vllm/v1/metrics/reader.py +257 -0
- vllm/v1/metrics/stats.py +420 -0
- vllm/v1/outputs.py +249 -0
- vllm/v1/pool/__init__.py +0 -0
- vllm/v1/pool/metadata.py +82 -0
- vllm/v1/request.py +259 -0
- vllm/v1/sample/__init__.py +0 -0
- vllm/v1/sample/logits_processor/__init__.py +352 -0
- vllm/v1/sample/logits_processor/builtin.py +274 -0
- vllm/v1/sample/logits_processor/interface.py +106 -0
- vllm/v1/sample/logits_processor/state.py +165 -0
- vllm/v1/sample/metadata.py +44 -0
- vllm/v1/sample/ops/__init__.py +0 -0
- vllm/v1/sample/ops/bad_words.py +52 -0
- vllm/v1/sample/ops/logprobs.py +25 -0
- vllm/v1/sample/ops/penalties.py +57 -0
- vllm/v1/sample/ops/topk_topp_sampler.py +290 -0
- vllm/v1/sample/rejection_sampler.py +793 -0
- vllm/v1/sample/sampler.py +316 -0
- vllm/v1/sample/tpu/__init__.py +0 -0
- vllm/v1/sample/tpu/metadata.py +120 -0
- vllm/v1/sample/tpu/sampler.py +215 -0
- vllm/v1/serial_utils.py +532 -0
- vllm/v1/spec_decode/__init__.py +0 -0
- vllm/v1/spec_decode/eagle.py +1225 -0
- vllm/v1/spec_decode/medusa.py +73 -0
- vllm/v1/spec_decode/metadata.py +66 -0
- vllm/v1/spec_decode/metrics.py +224 -0
- vllm/v1/spec_decode/ngram_proposer.py +291 -0
- vllm/v1/spec_decode/suffix_decoding.py +103 -0
- vllm/v1/spec_decode/utils.py +16 -0
- vllm/v1/structured_output/__init__.py +338 -0
- vllm/v1/structured_output/backend_guidance.py +265 -0
- vllm/v1/structured_output/backend_lm_format_enforcer.py +177 -0
- vllm/v1/structured_output/backend_outlines.py +324 -0
- vllm/v1/structured_output/backend_types.py +136 -0
- vllm/v1/structured_output/backend_xgrammar.py +362 -0
- vllm/v1/structured_output/request.py +94 -0
- vllm/v1/structured_output/utils.py +469 -0
- vllm/v1/utils.py +414 -0
- vllm/v1/worker/__init__.py +0 -0
- vllm/v1/worker/block_table.py +327 -0
- vllm/v1/worker/cpu_model_runner.py +122 -0
- vllm/v1/worker/cpu_worker.py +206 -0
- vllm/v1/worker/dp_utils.py +230 -0
- vllm/v1/worker/ec_connector_model_runner_mixin.py +87 -0
- vllm/v1/worker/gpu_input_batch.py +975 -0
- vllm/v1/worker/gpu_model_runner.py +5102 -0
- vllm/v1/worker/gpu_ubatch_wrapper.py +466 -0
- vllm/v1/worker/gpu_worker.py +894 -0
- vllm/v1/worker/kv_connector_model_runner_mixin.py +144 -0
- vllm/v1/worker/lora_model_runner_mixin.py +213 -0
- vllm/v1/worker/tpu_input_batch.py +593 -0
- vllm/v1/worker/tpu_model_runner.py +2173 -0
- vllm/v1/worker/tpu_worker.py +355 -0
- vllm/v1/worker/ubatch_utils.py +73 -0
- vllm/v1/worker/ubatching.py +231 -0
- vllm/v1/worker/utils.py +366 -0
- vllm/v1/worker/worker_base.py +375 -0
- vllm/v1/worker/xpu_model_runner.py +55 -0
- vllm/v1/worker/xpu_worker.py +189 -0
- vllm/version.py +39 -0
- vllm/vllm_flash_attn/.gitkeep +0 -0
- vllm_cpu_amxbf16-0.11.2.post2.dist-info/METADATA +345 -0
- vllm_cpu_amxbf16-0.11.2.post2.dist-info/RECORD +1536 -0
- vllm_cpu_amxbf16-0.11.2.post2.dist-info/WHEEL +5 -0
- vllm_cpu_amxbf16-0.11.2.post2.dist-info/entry_points.txt +5 -0
- vllm_cpu_amxbf16-0.11.2.post2.dist-info/top_level.txt +1 -0
|
@@ -0,0 +1,744 @@
|
|
|
1
|
+
# SPDX-License-Identifier: Apache-2.0
|
|
2
|
+
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
3
|
+
# Adapted from qwen3coder xml parser, All rights reserved.
|
|
4
|
+
# ruff: noqa: E501
|
|
5
|
+
|
|
6
|
+
import ast
|
|
7
|
+
import json
|
|
8
|
+
import uuid
|
|
9
|
+
from collections.abc import Sequence
|
|
10
|
+
from typing import Any
|
|
11
|
+
|
|
12
|
+
import regex as re
|
|
13
|
+
|
|
14
|
+
from vllm.entrypoints.openai.protocol import (
|
|
15
|
+
ChatCompletionRequest,
|
|
16
|
+
ChatCompletionToolsParam,
|
|
17
|
+
DeltaFunctionCall,
|
|
18
|
+
DeltaMessage,
|
|
19
|
+
DeltaToolCall,
|
|
20
|
+
ExtractedToolCallInformation,
|
|
21
|
+
FunctionCall,
|
|
22
|
+
ToolCall,
|
|
23
|
+
)
|
|
24
|
+
from vllm.entrypoints.openai.tool_parsers.abstract_tool_parser import (
|
|
25
|
+
ToolParser,
|
|
26
|
+
)
|
|
27
|
+
from vllm.logger import init_logger
|
|
28
|
+
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
|
29
|
+
|
|
30
|
+
logger = init_logger(__name__)
|
|
31
|
+
|
|
32
|
+
|
|
33
|
+
class SeedOssToolParser(ToolParser):
|
|
34
|
+
TOOL_CALL_START = "<seed:tool_call>"
|
|
35
|
+
TOOL_CALL_END = "</seed:tool_call>"
|
|
36
|
+
|
|
37
|
+
def __init__(self, tokenizer: AnyTokenizer):
|
|
38
|
+
super().__init__(tokenizer)
|
|
39
|
+
|
|
40
|
+
# --- streaming state ---
|
|
41
|
+
self._reset_streaming_state()
|
|
42
|
+
self.prev_tool_call_arr: list[dict] = []
|
|
43
|
+
|
|
44
|
+
self.tool_call_start_token: str = self.TOOL_CALL_START
|
|
45
|
+
self.tool_call_end_token: str = self.TOOL_CALL_END
|
|
46
|
+
# Sentinel tokens for streaming mode
|
|
47
|
+
self.tool_call_prefix: str = "<function="
|
|
48
|
+
self.function_end_token: str = "</function>"
|
|
49
|
+
self.parameter_prefix: str = "<parameter="
|
|
50
|
+
self.parameter_end_token: str = "</parameter>"
|
|
51
|
+
self.think_start_token: str = "<seed:think>"
|
|
52
|
+
self.think_end_token: str = "</seed:think>"
|
|
53
|
+
self.is_tool_call_started: bool = False
|
|
54
|
+
self.is_thinking_end: bool = False
|
|
55
|
+
self.failed_count: int = 0
|
|
56
|
+
self._reset_streaming_state()
|
|
57
|
+
|
|
58
|
+
self.tool_call_start_token_id = self.vocab.get(self.tool_call_start_token)
|
|
59
|
+
self.tool_call_end_token_id = self.vocab.get(self.tool_call_end_token)
|
|
60
|
+
self.think_end_token_id = self.vocab.get(self.think_end_token)
|
|
61
|
+
|
|
62
|
+
if self.tool_call_start_token_id is None or self.tool_call_end_token_id is None:
|
|
63
|
+
raise RuntimeError(
|
|
64
|
+
"Seed_Oss XML parser: tokenizer did not include "
|
|
65
|
+
"<seed:tool_call> or its closing tag."
|
|
66
|
+
)
|
|
67
|
+
|
|
68
|
+
tool_start_re = re.escape(self.tool_call_start_token)
|
|
69
|
+
tool_end_re = re.escape(self.tool_call_end_token)
|
|
70
|
+
|
|
71
|
+
self.tool_call_complete_regex = re.compile(
|
|
72
|
+
rf"{tool_start_re}(.*?){tool_end_re}", re.DOTALL
|
|
73
|
+
)
|
|
74
|
+
self.tool_call_regex = re.compile(
|
|
75
|
+
rf"{tool_start_re}(.*?){tool_end_re}|{tool_start_re}(.*?)$", re.DOTALL
|
|
76
|
+
)
|
|
77
|
+
|
|
78
|
+
self.tool_call_function_regex = re.compile(
|
|
79
|
+
r"<function=(.*?)</function>|<function=(.*)$", re.DOTALL
|
|
80
|
+
)
|
|
81
|
+
self.tool_call_parameter_regex = re.compile(
|
|
82
|
+
r"<parameter=(.*?)</parameter>|<parameter=(.*?)$", re.DOTALL
|
|
83
|
+
)
|
|
84
|
+
|
|
85
|
+
logger.info(
|
|
86
|
+
"vLLM Seed-Oss XML tool parser loaded (%s).", self.__class__.__name__
|
|
87
|
+
)
|
|
88
|
+
|
|
89
|
+
def _generate_tool_call_id(self) -> str:
|
|
90
|
+
"""Generate a unique tool call ID."""
|
|
91
|
+
return f"call_{uuid.uuid4().hex[:24]}"
|
|
92
|
+
|
|
93
|
+
def _reset_streaming_state(self):
|
|
94
|
+
"""Reset all streaming state."""
|
|
95
|
+
self.current_tool_index = 0
|
|
96
|
+
self.is_tool_call_started = False
|
|
97
|
+
self.header_sent = False
|
|
98
|
+
self.current_tool_id = -1
|
|
99
|
+
self.current_function_name = None
|
|
100
|
+
self.current_param_name = None
|
|
101
|
+
self.current_param_value = ""
|
|
102
|
+
self.param_count = 0
|
|
103
|
+
self.in_param = False
|
|
104
|
+
self.in_function = False
|
|
105
|
+
self.accumulated_text = ""
|
|
106
|
+
self.json_started = False
|
|
107
|
+
self.json_closed = False
|
|
108
|
+
|
|
109
|
+
def _parse_xml_function_call(
|
|
110
|
+
self, function_call_str: str, tools: list[ChatCompletionToolsParam] | None
|
|
111
|
+
) -> ToolCall | None:
|
|
112
|
+
def get_arguments_config(func_name: str) -> dict:
|
|
113
|
+
if tools is None:
|
|
114
|
+
return {}
|
|
115
|
+
for config in tools:
|
|
116
|
+
if not hasattr(config, "type") or not (
|
|
117
|
+
hasattr(config, "function") and hasattr(config.function, "name")
|
|
118
|
+
):
|
|
119
|
+
continue
|
|
120
|
+
if config.type == "function" and config.function.name == func_name:
|
|
121
|
+
if not hasattr(config.function, "parameters"):
|
|
122
|
+
return {}
|
|
123
|
+
params = config.function.parameters
|
|
124
|
+
if isinstance(params, dict) and "properties" in params:
|
|
125
|
+
return params["properties"]
|
|
126
|
+
elif isinstance(params, dict):
|
|
127
|
+
return params
|
|
128
|
+
else:
|
|
129
|
+
return {}
|
|
130
|
+
logger.warning("Tool '%s' is not defined in the tools list.", func_name)
|
|
131
|
+
return {}
|
|
132
|
+
|
|
133
|
+
def convert_param_value(
|
|
134
|
+
param_value: str, param_name: str, param_config: dict, func_name: str
|
|
135
|
+
) -> Any:
|
|
136
|
+
# Handle null value for any type
|
|
137
|
+
if param_value.lower() == "null":
|
|
138
|
+
return None
|
|
139
|
+
|
|
140
|
+
if param_name not in param_config:
|
|
141
|
+
if param_config != {}:
|
|
142
|
+
logger.warning(
|
|
143
|
+
"Parsed parameter '%s' is not defined in "
|
|
144
|
+
"the tool parameters for tool '%s', "
|
|
145
|
+
"directly returning the string value.",
|
|
146
|
+
param_name,
|
|
147
|
+
func_name,
|
|
148
|
+
)
|
|
149
|
+
return param_value
|
|
150
|
+
|
|
151
|
+
if (
|
|
152
|
+
isinstance(param_config[param_name], dict)
|
|
153
|
+
and "type" in param_config[param_name]
|
|
154
|
+
):
|
|
155
|
+
param_type = str(param_config[param_name]["type"]).strip().lower()
|
|
156
|
+
else:
|
|
157
|
+
param_type = "string"
|
|
158
|
+
if param_type in ["string", "str", "text", "varchar", "char", "enum"]:
|
|
159
|
+
return param_value
|
|
160
|
+
elif (
|
|
161
|
+
param_type.startswith("int")
|
|
162
|
+
or param_type.startswith("uint")
|
|
163
|
+
or param_type.startswith("long")
|
|
164
|
+
or param_type.startswith("short")
|
|
165
|
+
or param_type.startswith("unsigned")
|
|
166
|
+
):
|
|
167
|
+
try:
|
|
168
|
+
param_value = int(param_value) # type: ignore
|
|
169
|
+
except (ValueError, TypeError):
|
|
170
|
+
logger.warning(
|
|
171
|
+
"Parsed value '%s' of parameter '%s' is not an integer in tool "
|
|
172
|
+
"'%s', degenerating to string.",
|
|
173
|
+
param_value,
|
|
174
|
+
param_name,
|
|
175
|
+
func_name,
|
|
176
|
+
)
|
|
177
|
+
return param_value
|
|
178
|
+
elif param_type.startswith("num") or param_type.startswith("float"):
|
|
179
|
+
try:
|
|
180
|
+
float_param_value = float(param_value)
|
|
181
|
+
param_value = (
|
|
182
|
+
float_param_value # type: ignore
|
|
183
|
+
if float_param_value - int(float_param_value) != 0
|
|
184
|
+
else int(float_param_value) # type: ignore
|
|
185
|
+
)
|
|
186
|
+
except (ValueError, TypeError):
|
|
187
|
+
logger.warning(
|
|
188
|
+
"Parsed value '%s' of parameter '%s' is not a float in tool "
|
|
189
|
+
"'%s', degenerating to string.",
|
|
190
|
+
param_value,
|
|
191
|
+
param_name,
|
|
192
|
+
func_name,
|
|
193
|
+
)
|
|
194
|
+
return param_value
|
|
195
|
+
elif param_type in ["boolean", "bool", "binary"]:
|
|
196
|
+
param_value = param_value.lower()
|
|
197
|
+
if param_value not in ["true", "false"]:
|
|
198
|
+
logger.warning(
|
|
199
|
+
"Parsed value '%s' of parameter '%s' is not a boolean "
|
|
200
|
+
"(`true` of `false`) in tool '%s', degenerating to false.",
|
|
201
|
+
param_value,
|
|
202
|
+
param_name,
|
|
203
|
+
func_name,
|
|
204
|
+
)
|
|
205
|
+
return param_value == "true"
|
|
206
|
+
else:
|
|
207
|
+
if param_type == "object" or param_type.startswith("dict"):
|
|
208
|
+
try:
|
|
209
|
+
param_value = json.loads(param_value)
|
|
210
|
+
return param_value
|
|
211
|
+
except (ValueError, TypeError, json.JSONDecodeError):
|
|
212
|
+
logger.warning(
|
|
213
|
+
"Parsed value '%s' of parameter '%s' is not a valid JSON "
|
|
214
|
+
"object in tool '%s', will try other methods to parse it.",
|
|
215
|
+
param_value,
|
|
216
|
+
param_name,
|
|
217
|
+
func_name,
|
|
218
|
+
)
|
|
219
|
+
try:
|
|
220
|
+
param_value = ast.literal_eval(param_value)
|
|
221
|
+
except (ValueError, SyntaxError):
|
|
222
|
+
logger.warning(
|
|
223
|
+
"Parsed value '%s' of parameter '%s' cannot be converted via "
|
|
224
|
+
"Python `ast.literal_eval()` in tool '%s', degenerating to string.",
|
|
225
|
+
param_value,
|
|
226
|
+
param_name,
|
|
227
|
+
func_name,
|
|
228
|
+
)
|
|
229
|
+
return param_value
|
|
230
|
+
|
|
231
|
+
# Extract function name
|
|
232
|
+
end_index = function_call_str.index(">")
|
|
233
|
+
function_name = function_call_str[:end_index]
|
|
234
|
+
param_config = get_arguments_config(function_name)
|
|
235
|
+
parameters = function_call_str[end_index + 1 :]
|
|
236
|
+
param_dict = {}
|
|
237
|
+
for match in self.tool_call_parameter_regex.findall(parameters):
|
|
238
|
+
match_text = match[0] if match[0] else match[1]
|
|
239
|
+
idx = match_text.index(">")
|
|
240
|
+
param_name = match_text[:idx]
|
|
241
|
+
param_value = str(match_text[idx + 1 :])
|
|
242
|
+
# Remove prefix and trailing \n
|
|
243
|
+
if param_value.startswith("\n"):
|
|
244
|
+
param_value = param_value[1:]
|
|
245
|
+
if param_value.endswith("\n"):
|
|
246
|
+
param_value = param_value[:-1]
|
|
247
|
+
|
|
248
|
+
param_dict[param_name] = convert_param_value(
|
|
249
|
+
param_value, param_name, param_config, function_name
|
|
250
|
+
)
|
|
251
|
+
return ToolCall(
|
|
252
|
+
type="function",
|
|
253
|
+
function=FunctionCall(
|
|
254
|
+
name=function_name, arguments=json.dumps(param_dict, ensure_ascii=False)
|
|
255
|
+
),
|
|
256
|
+
)
|
|
257
|
+
|
|
258
|
+
def _get_function_calls(self, model_output: str) -> list[str]:
|
|
259
|
+
# Find all tool calls
|
|
260
|
+
matched_ranges = self.tool_call_regex.findall(model_output)
|
|
261
|
+
raw_tool_calls = [
|
|
262
|
+
match[0] if match[0] else match[1] for match in matched_ranges
|
|
263
|
+
]
|
|
264
|
+
|
|
265
|
+
# Back-off strategy if no tool_call tags found
|
|
266
|
+
if len(raw_tool_calls) == 0:
|
|
267
|
+
raw_tool_calls = [model_output]
|
|
268
|
+
|
|
269
|
+
raw_function_calls = []
|
|
270
|
+
for tool_call in raw_tool_calls:
|
|
271
|
+
raw_function_calls.extend(self.tool_call_function_regex.findall(tool_call))
|
|
272
|
+
|
|
273
|
+
function_calls = [
|
|
274
|
+
match[0] if match[0] else match[1] for match in raw_function_calls
|
|
275
|
+
]
|
|
276
|
+
return function_calls
|
|
277
|
+
|
|
278
|
+
def extract_tool_calls(
|
|
279
|
+
self,
|
|
280
|
+
model_output: str,
|
|
281
|
+
request: ChatCompletionRequest,
|
|
282
|
+
) -> ExtractedToolCallInformation:
|
|
283
|
+
# Quick check to avoid unnecessary processing
|
|
284
|
+
if self.tool_call_prefix not in model_output:
|
|
285
|
+
return ExtractedToolCallInformation(
|
|
286
|
+
tools_called=False, tool_calls=[], content=model_output
|
|
287
|
+
)
|
|
288
|
+
|
|
289
|
+
# Check if both think start and end tokens are present
|
|
290
|
+
if (
|
|
291
|
+
self.think_start_token in model_output
|
|
292
|
+
and self.think_end_token in model_output
|
|
293
|
+
):
|
|
294
|
+
# Find the position of think end token
|
|
295
|
+
think_end_index = model_output.find(self.think_end_token) + len(
|
|
296
|
+
self.think_end_token
|
|
297
|
+
)
|
|
298
|
+
# Extract content after think end token
|
|
299
|
+
result_content = model_output[think_end_index:]
|
|
300
|
+
thinking_content = model_output[:think_end_index]
|
|
301
|
+
else:
|
|
302
|
+
thinking_content = ""
|
|
303
|
+
result_content = model_output
|
|
304
|
+
|
|
305
|
+
try:
|
|
306
|
+
function_calls = self._get_function_calls(result_content)
|
|
307
|
+
if len(function_calls) == 0:
|
|
308
|
+
return ExtractedToolCallInformation(
|
|
309
|
+
tools_called=False, tool_calls=[], content=model_output
|
|
310
|
+
)
|
|
311
|
+
|
|
312
|
+
tool_calls = [
|
|
313
|
+
self._parse_xml_function_call(function_call_str, request.tools)
|
|
314
|
+
for function_call_str in function_calls
|
|
315
|
+
]
|
|
316
|
+
|
|
317
|
+
# Populate prev_tool_call_arr for serving layer to set finish_reason
|
|
318
|
+
self.prev_tool_call_arr.clear() # Clear previous calls
|
|
319
|
+
for tool_call in tool_calls:
|
|
320
|
+
if tool_call:
|
|
321
|
+
self.prev_tool_call_arr.append(
|
|
322
|
+
{
|
|
323
|
+
"name": tool_call.function.name,
|
|
324
|
+
"arguments": tool_call.function.arguments,
|
|
325
|
+
}
|
|
326
|
+
)
|
|
327
|
+
|
|
328
|
+
# Extract content before tool calls
|
|
329
|
+
tool_call_start_index = result_content.find(self.tool_call_start_token)
|
|
330
|
+
tool_call_start_index = (
|
|
331
|
+
tool_call_start_index
|
|
332
|
+
if tool_call_start_index >= 0
|
|
333
|
+
else result_content.find(self.tool_call_prefix)
|
|
334
|
+
)
|
|
335
|
+
content = thinking_content + result_content[:tool_call_start_index]
|
|
336
|
+
|
|
337
|
+
return ExtractedToolCallInformation(
|
|
338
|
+
tools_called=(len(tool_calls) > 0),
|
|
339
|
+
tool_calls=tool_calls,
|
|
340
|
+
content=content if content else None,
|
|
341
|
+
)
|
|
342
|
+
|
|
343
|
+
except Exception:
|
|
344
|
+
logger.exception("Error in extracting tool call from response.")
|
|
345
|
+
return ExtractedToolCallInformation(
|
|
346
|
+
tools_called=False, tool_calls=[], content=model_output
|
|
347
|
+
)
|
|
348
|
+
|
|
349
|
+
def extract_tool_calls_streaming(
|
|
350
|
+
self,
|
|
351
|
+
previous_text: str,
|
|
352
|
+
current_text: str,
|
|
353
|
+
delta_text: str,
|
|
354
|
+
previous_token_ids: Sequence[int],
|
|
355
|
+
current_token_ids: Sequence[int],
|
|
356
|
+
delta_token_ids: Sequence[int],
|
|
357
|
+
request: ChatCompletionRequest,
|
|
358
|
+
) -> DeltaMessage | None:
|
|
359
|
+
# If no delta text, return None unless
|
|
360
|
+
# it's an EOS token after tool calls
|
|
361
|
+
if not delta_text:
|
|
362
|
+
# Check if this is an EOS token after all tool calls are complete
|
|
363
|
+
# We check for tool calls in the text even if is_tool_call_started
|
|
364
|
+
# is False because it might have been reset after processing all tools
|
|
365
|
+
if delta_token_ids and self.tool_call_end_token_id not in delta_token_ids:
|
|
366
|
+
# Count complete tool calls
|
|
367
|
+
complete_calls = len(
|
|
368
|
+
self.tool_call_complete_regex.findall(current_text)
|
|
369
|
+
)
|
|
370
|
+
|
|
371
|
+
# If we have completed tool calls and populated prev_tool_call_arr
|
|
372
|
+
if complete_calls > 0 and len(self.prev_tool_call_arr) > 0:
|
|
373
|
+
# Check if all tool calls are closed
|
|
374
|
+
open_calls = current_text.count(
|
|
375
|
+
self.tool_call_start_token
|
|
376
|
+
) - current_text.count(self.tool_call_end_token)
|
|
377
|
+
if open_calls == 0:
|
|
378
|
+
# Return empty delta message to allow finish_reason processing
|
|
379
|
+
return DeltaMessage(content="")
|
|
380
|
+
elif not self.is_tool_call_started and current_text:
|
|
381
|
+
# This is a regular content response that's now complete
|
|
382
|
+
return DeltaMessage(content="")
|
|
383
|
+
return None
|
|
384
|
+
|
|
385
|
+
# Check if this is the first call (reset state if needed)
|
|
386
|
+
if not previous_text:
|
|
387
|
+
self._reset_streaming_state()
|
|
388
|
+
|
|
389
|
+
# Update accumulated text
|
|
390
|
+
self.accumulated_text = current_text
|
|
391
|
+
|
|
392
|
+
# Check if we need to advance to next tool
|
|
393
|
+
if self.json_closed and not self.in_function:
|
|
394
|
+
# Check if this tool call has ended
|
|
395
|
+
tool_ends = current_text.count(self.tool_call_end_token)
|
|
396
|
+
if tool_ends > self.current_tool_index:
|
|
397
|
+
# This tool has ended, advance to next
|
|
398
|
+
self.current_tool_index += 1
|
|
399
|
+
self.header_sent = False
|
|
400
|
+
self.param_count = 0
|
|
401
|
+
self.json_started = False
|
|
402
|
+
self.json_closed = False
|
|
403
|
+
|
|
404
|
+
# Check if there are more tool calls
|
|
405
|
+
if self.current_tool_index >= current_text.count(
|
|
406
|
+
self.tool_call_start_token
|
|
407
|
+
):
|
|
408
|
+
# No more tool calls
|
|
409
|
+
self.is_tool_call_started = False
|
|
410
|
+
# Continue processing next tool
|
|
411
|
+
return None
|
|
412
|
+
|
|
413
|
+
# Check if end thinking
|
|
414
|
+
if not self.is_thinking_end and (
|
|
415
|
+
self.think_end_token_id in delta_token_ids
|
|
416
|
+
or self.think_end_token in delta_text
|
|
417
|
+
):
|
|
418
|
+
self.is_thinking_end = True
|
|
419
|
+
|
|
420
|
+
# If thinking hasn't ended yet, don't process any tool calls
|
|
421
|
+
if not self.is_thinking_end:
|
|
422
|
+
return DeltaMessage(content=delta_text)
|
|
423
|
+
|
|
424
|
+
# Handle normal content before tool calls
|
|
425
|
+
if not self.is_tool_call_started:
|
|
426
|
+
# Check if tool call is starting
|
|
427
|
+
if (
|
|
428
|
+
self.tool_call_start_token_id in delta_token_ids
|
|
429
|
+
or self.tool_call_start_token in delta_text
|
|
430
|
+
):
|
|
431
|
+
self.is_tool_call_started = True
|
|
432
|
+
# Return any content before the tool call
|
|
433
|
+
if self.tool_call_start_token in delta_text:
|
|
434
|
+
content_before = delta_text[
|
|
435
|
+
: delta_text.index(self.tool_call_start_token)
|
|
436
|
+
]
|
|
437
|
+
if content_before:
|
|
438
|
+
return DeltaMessage(content=content_before)
|
|
439
|
+
return None
|
|
440
|
+
else:
|
|
441
|
+
# Check if we're between tool calls - skip whitespace
|
|
442
|
+
if (
|
|
443
|
+
current_text.rstrip().endswith(self.tool_call_end_token)
|
|
444
|
+
and delta_text.strip() == ""
|
|
445
|
+
):
|
|
446
|
+
# We just ended a tool call, skip whitespace
|
|
447
|
+
return None
|
|
448
|
+
# Normal content, no tool call
|
|
449
|
+
return DeltaMessage(content=delta_text)
|
|
450
|
+
|
|
451
|
+
# Check if we're between tool calls (waiting for next one)
|
|
452
|
+
# Count tool calls we've seen vs processed
|
|
453
|
+
tool_starts_count = current_text.count(self.tool_call_start_token)
|
|
454
|
+
if self.current_tool_index >= tool_starts_count:
|
|
455
|
+
# We're past all tool calls, shouldn't be here
|
|
456
|
+
return None
|
|
457
|
+
|
|
458
|
+
# We're in a tool call, find the current tool call portion
|
|
459
|
+
# Need to find the correct tool call based on current_tool_index
|
|
460
|
+
# Only process tool calls after think_end_token
|
|
461
|
+
think_end_index = (
|
|
462
|
+
current_text.find(self.think_end_token) + len(self.think_end_token)
|
|
463
|
+
if self.think_end_token in current_text
|
|
464
|
+
else 0
|
|
465
|
+
)
|
|
466
|
+
tool_starts: list[int] = []
|
|
467
|
+
idx = think_end_index
|
|
468
|
+
while True:
|
|
469
|
+
idx = current_text.find(self.tool_call_start_token, idx)
|
|
470
|
+
if idx == -1:
|
|
471
|
+
break
|
|
472
|
+
tool_starts.append(idx)
|
|
473
|
+
idx += len(self.tool_call_start_token)
|
|
474
|
+
|
|
475
|
+
if self.current_tool_index >= len(tool_starts):
|
|
476
|
+
# No more tool calls to process yet
|
|
477
|
+
return None
|
|
478
|
+
|
|
479
|
+
tool_start_idx = tool_starts[self.current_tool_index]
|
|
480
|
+
# Find where this tool call ends (or current position if not ended yet)
|
|
481
|
+
tool_end_idx = current_text.find(self.tool_call_end_token, tool_start_idx)
|
|
482
|
+
if tool_end_idx == -1:
|
|
483
|
+
tool_text = current_text[tool_start_idx:]
|
|
484
|
+
else:
|
|
485
|
+
tool_text = current_text[
|
|
486
|
+
tool_start_idx : tool_end_idx + len(self.tool_call_end_token)
|
|
487
|
+
]
|
|
488
|
+
|
|
489
|
+
# Looking for function header
|
|
490
|
+
if not self.header_sent:
|
|
491
|
+
if self.tool_call_prefix in tool_text:
|
|
492
|
+
func_start = tool_text.find(self.tool_call_prefix) + len(
|
|
493
|
+
self.tool_call_prefix
|
|
494
|
+
)
|
|
495
|
+
func_end = tool_text.find(">", func_start)
|
|
496
|
+
|
|
497
|
+
if func_end != -1:
|
|
498
|
+
# Found complete function name
|
|
499
|
+
self.current_function_name = tool_text[func_start:func_end]
|
|
500
|
+
self.current_tool_id = self._generate_tool_call_id() # type: ignore
|
|
501
|
+
self.header_sent = True
|
|
502
|
+
self.in_function = True
|
|
503
|
+
|
|
504
|
+
# IMPORTANT: Add to prev_tool_call_arr immediately when we detect a tool call
|
|
505
|
+
# This ensures finish_reason="tool_calls" even if parsing isn't complete
|
|
506
|
+
already_added = any(
|
|
507
|
+
tool.get("name") == self.current_function_name
|
|
508
|
+
for tool in self.prev_tool_call_arr
|
|
509
|
+
)
|
|
510
|
+
if not already_added:
|
|
511
|
+
self.prev_tool_call_arr.append(
|
|
512
|
+
{
|
|
513
|
+
"name": self.current_function_name,
|
|
514
|
+
"arguments": "{}", # Placeholder, will be updated later
|
|
515
|
+
}
|
|
516
|
+
)
|
|
517
|
+
|
|
518
|
+
# Send header with function info
|
|
519
|
+
return DeltaMessage(
|
|
520
|
+
tool_calls=[
|
|
521
|
+
DeltaToolCall(
|
|
522
|
+
index=self.current_tool_index,
|
|
523
|
+
id=self.current_tool_id,
|
|
524
|
+
function=DeltaFunctionCall(
|
|
525
|
+
name=self.current_function_name, arguments=""
|
|
526
|
+
),
|
|
527
|
+
type="function",
|
|
528
|
+
)
|
|
529
|
+
]
|
|
530
|
+
)
|
|
531
|
+
return None
|
|
532
|
+
|
|
533
|
+
# We've sent header, now handle function body
|
|
534
|
+
if self.in_function:
|
|
535
|
+
# Send opening brace if not sent yet
|
|
536
|
+
if not self.json_started and self.parameter_prefix not in delta_text:
|
|
537
|
+
self.json_started = True
|
|
538
|
+
return DeltaMessage(
|
|
539
|
+
tool_calls=[
|
|
540
|
+
DeltaToolCall(
|
|
541
|
+
index=self.current_tool_index,
|
|
542
|
+
function=DeltaFunctionCall(arguments="{"),
|
|
543
|
+
)
|
|
544
|
+
]
|
|
545
|
+
)
|
|
546
|
+
|
|
547
|
+
# Make sure json_started is set if we're processing parameters
|
|
548
|
+
if not self.json_started:
|
|
549
|
+
self.json_started = True
|
|
550
|
+
|
|
551
|
+
# Check for function end in accumulated text
|
|
552
|
+
if not self.json_closed and self.function_end_token in tool_text:
|
|
553
|
+
# Close JSON
|
|
554
|
+
self.json_closed = True
|
|
555
|
+
|
|
556
|
+
# Extract the complete tool call to update prev_tool_call_arr with final arguments
|
|
557
|
+
# Find the function content
|
|
558
|
+
func_start = tool_text.find(self.tool_call_prefix) + len(
|
|
559
|
+
self.tool_call_prefix
|
|
560
|
+
)
|
|
561
|
+
func_content_end = tool_text.find(self.function_end_token, func_start)
|
|
562
|
+
if func_content_end != -1:
|
|
563
|
+
func_content = tool_text[func_start:func_content_end]
|
|
564
|
+
# Parse to get the complete arguments
|
|
565
|
+
try:
|
|
566
|
+
parsed_tool = self._parse_xml_function_call(
|
|
567
|
+
func_content, request.tools if request else None
|
|
568
|
+
)
|
|
569
|
+
if parsed_tool:
|
|
570
|
+
# Update existing entry in prev_tool_call_arr with complete arguments
|
|
571
|
+
for i, tool in enumerate(self.prev_tool_call_arr):
|
|
572
|
+
if tool.get("name") == parsed_tool.function.name:
|
|
573
|
+
self.prev_tool_call_arr[i]["arguments"] = (
|
|
574
|
+
parsed_tool.function.arguments
|
|
575
|
+
)
|
|
576
|
+
break
|
|
577
|
+
except Exception:
|
|
578
|
+
logger.warning(
|
|
579
|
+
"Failed to parse tool arguments during streaming.",
|
|
580
|
+
exc_info=True,
|
|
581
|
+
)
|
|
582
|
+
|
|
583
|
+
result = DeltaMessage(
|
|
584
|
+
tool_calls=[
|
|
585
|
+
DeltaToolCall(
|
|
586
|
+
index=self.current_tool_index,
|
|
587
|
+
function=DeltaFunctionCall(arguments="}"),
|
|
588
|
+
)
|
|
589
|
+
]
|
|
590
|
+
)
|
|
591
|
+
|
|
592
|
+
# Reset state for next tool
|
|
593
|
+
self.in_function = False
|
|
594
|
+
self.json_closed = True
|
|
595
|
+
|
|
596
|
+
return result
|
|
597
|
+
|
|
598
|
+
# Look for parameters
|
|
599
|
+
# Count how many complete parameters we have processed
|
|
600
|
+
complete_params = tool_text.count(self.parameter_end_token)
|
|
601
|
+
|
|
602
|
+
# Check if we should start a new parameter
|
|
603
|
+
if not self.in_param and self.param_count < complete_params:
|
|
604
|
+
# Find the unprocessed parameter
|
|
605
|
+
# Count parameter starts
|
|
606
|
+
param_starts = []
|
|
607
|
+
idx = 0
|
|
608
|
+
while True:
|
|
609
|
+
idx = tool_text.find(self.parameter_prefix, idx)
|
|
610
|
+
if idx == -1:
|
|
611
|
+
break
|
|
612
|
+
param_starts.append(idx)
|
|
613
|
+
idx += len(self.parameter_prefix)
|
|
614
|
+
|
|
615
|
+
if len(param_starts) > self.param_count:
|
|
616
|
+
# Process the next parameter
|
|
617
|
+
param_idx = param_starts[self.param_count]
|
|
618
|
+
param_start = param_idx + len(self.parameter_prefix)
|
|
619
|
+
remaining = tool_text[param_start:]
|
|
620
|
+
|
|
621
|
+
if ">" in remaining:
|
|
622
|
+
# We have the complete parameter name
|
|
623
|
+
name_end = remaining.find(">")
|
|
624
|
+
self.current_param_name = remaining[:name_end]
|
|
625
|
+
|
|
626
|
+
# Find the parameter value
|
|
627
|
+
value_start = param_start + name_end + 1
|
|
628
|
+
value_text = tool_text[value_start:]
|
|
629
|
+
if value_text.startswith("\n"):
|
|
630
|
+
value_text = value_text[1:]
|
|
631
|
+
|
|
632
|
+
# Find where this parameter ends
|
|
633
|
+
param_end_idx = value_text.find(self.parameter_end_token)
|
|
634
|
+
if param_end_idx != -1:
|
|
635
|
+
# Complete parameter found
|
|
636
|
+
param_value = value_text[:param_end_idx]
|
|
637
|
+
if param_value.endswith("\n"):
|
|
638
|
+
param_value = param_value[:-1]
|
|
639
|
+
|
|
640
|
+
# Build complete JSON fragment for this parameter
|
|
641
|
+
if self.param_count == 0:
|
|
642
|
+
json_fragment = (
|
|
643
|
+
'"'
|
|
644
|
+
+ self.current_param_name
|
|
645
|
+
+ '": "'
|
|
646
|
+
+ json.dumps(param_value)[1:-1]
|
|
647
|
+
+ '"'
|
|
648
|
+
)
|
|
649
|
+
else:
|
|
650
|
+
json_fragment = (
|
|
651
|
+
', "'
|
|
652
|
+
+ self.current_param_name
|
|
653
|
+
+ '": "'
|
|
654
|
+
+ json.dumps(param_value)[1:-1]
|
|
655
|
+
+ '"'
|
|
656
|
+
)
|
|
657
|
+
|
|
658
|
+
self.param_count += 1
|
|
659
|
+
|
|
660
|
+
return DeltaMessage(
|
|
661
|
+
tool_calls=[
|
|
662
|
+
DeltaToolCall(
|
|
663
|
+
index=self.current_tool_index,
|
|
664
|
+
function=DeltaFunctionCall(
|
|
665
|
+
arguments=json_fragment
|
|
666
|
+
),
|
|
667
|
+
)
|
|
668
|
+
]
|
|
669
|
+
)
|
|
670
|
+
|
|
671
|
+
# Continue parameter value
|
|
672
|
+
if self.in_param:
|
|
673
|
+
if self.parameter_end_token in delta_text:
|
|
674
|
+
# End of parameter
|
|
675
|
+
end_idx = delta_text.find(self.parameter_end_token)
|
|
676
|
+
value_chunk = delta_text[:end_idx]
|
|
677
|
+
|
|
678
|
+
# Skip past > if at start
|
|
679
|
+
if not self.current_param_value and ">" in value_chunk:
|
|
680
|
+
gt_idx = value_chunk.find(">")
|
|
681
|
+
value_chunk = value_chunk[gt_idx + 1 :]
|
|
682
|
+
|
|
683
|
+
if not self.current_param_value and value_chunk.startswith("\n"):
|
|
684
|
+
value_chunk = value_chunk[1:]
|
|
685
|
+
|
|
686
|
+
# Calculate incremental JSON
|
|
687
|
+
full_value = self.current_param_value + value_chunk
|
|
688
|
+
prev_escaped = (
|
|
689
|
+
json.dumps(self.current_param_value)[1:-1]
|
|
690
|
+
if self.current_param_value
|
|
691
|
+
else ""
|
|
692
|
+
)
|
|
693
|
+
full_escaped = json.dumps(full_value)[1:-1]
|
|
694
|
+
delta_escaped = full_escaped[len(prev_escaped) :]
|
|
695
|
+
|
|
696
|
+
self.in_param = False
|
|
697
|
+
self.current_param_value = ""
|
|
698
|
+
|
|
699
|
+
return DeltaMessage(
|
|
700
|
+
tool_calls=[
|
|
701
|
+
DeltaToolCall(
|
|
702
|
+
index=self.current_tool_index,
|
|
703
|
+
function=DeltaFunctionCall(
|
|
704
|
+
arguments=delta_escaped + '"'
|
|
705
|
+
),
|
|
706
|
+
)
|
|
707
|
+
]
|
|
708
|
+
)
|
|
709
|
+
else:
|
|
710
|
+
# Continue accumulating value
|
|
711
|
+
value_chunk = delta_text
|
|
712
|
+
|
|
713
|
+
# Handle first chunk after param name
|
|
714
|
+
if not self.current_param_value and ">" in value_chunk:
|
|
715
|
+
gt_idx = value_chunk.find(">")
|
|
716
|
+
value_chunk = value_chunk[gt_idx + 1 :]
|
|
717
|
+
|
|
718
|
+
if not self.current_param_value and value_chunk.startswith("\n"):
|
|
719
|
+
value_chunk = value_chunk[1:]
|
|
720
|
+
|
|
721
|
+
if value_chunk:
|
|
722
|
+
# Stream the escaped delta
|
|
723
|
+
prev_escaped = (
|
|
724
|
+
json.dumps(self.current_param_value)[1:-1]
|
|
725
|
+
if self.current_param_value
|
|
726
|
+
else ""
|
|
727
|
+
)
|
|
728
|
+
self.current_param_value += value_chunk
|
|
729
|
+
full_escaped = json.dumps(self.current_param_value)[1:-1]
|
|
730
|
+
delta_escaped = full_escaped[len(prev_escaped) :]
|
|
731
|
+
|
|
732
|
+
if delta_escaped:
|
|
733
|
+
return DeltaMessage(
|
|
734
|
+
tool_calls=[
|
|
735
|
+
DeltaToolCall(
|
|
736
|
+
index=self.current_tool_index,
|
|
737
|
+
function=DeltaFunctionCall(
|
|
738
|
+
arguments=delta_escaped
|
|
739
|
+
),
|
|
740
|
+
)
|
|
741
|
+
]
|
|
742
|
+
)
|
|
743
|
+
|
|
744
|
+
return None
|