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.
Files changed (1536) hide show
  1. vllm/_C.abi3.so +0 -0
  2. vllm/__init__.py +225 -0
  3. vllm/_aiter_ops.py +983 -0
  4. vllm/_bc_linter.py +54 -0
  5. vllm/_custom_ops.py +2863 -0
  6. vllm/_ipex_ops.py +457 -0
  7. vllm/_version.py +34 -0
  8. vllm/assets/__init__.py +0 -0
  9. vllm/assets/audio.py +43 -0
  10. vllm/assets/base.py +40 -0
  11. vllm/assets/image.py +59 -0
  12. vllm/assets/video.py +149 -0
  13. vllm/attention/__init__.py +18 -0
  14. vllm/attention/backends/__init__.py +0 -0
  15. vllm/attention/backends/abstract.py +391 -0
  16. vllm/attention/backends/registry.py +195 -0
  17. vllm/attention/backends/utils.py +33 -0
  18. vllm/attention/layer.py +1052 -0
  19. vllm/attention/layers/__init__.py +0 -0
  20. vllm/attention/layers/chunked_local_attention.py +121 -0
  21. vllm/attention/layers/cross_attention.py +178 -0
  22. vllm/attention/layers/encoder_only_attention.py +103 -0
  23. vllm/attention/ops/__init__.py +0 -0
  24. vllm/attention/ops/chunked_prefill_paged_decode.py +401 -0
  25. vllm/attention/ops/common.py +414 -0
  26. vllm/attention/ops/flashmla.py +251 -0
  27. vllm/attention/ops/merge_attn_states.py +47 -0
  28. vllm/attention/ops/paged_attn.py +262 -0
  29. vllm/attention/ops/pallas_kv_cache_update.py +130 -0
  30. vllm/attention/ops/prefix_prefill.py +814 -0
  31. vllm/attention/ops/rocm_aiter_paged_attn.py +123 -0
  32. vllm/attention/ops/triton_decode_attention.py +712 -0
  33. vllm/attention/ops/triton_merge_attn_states.py +105 -0
  34. vllm/attention/ops/triton_reshape_and_cache_flash.py +184 -0
  35. vllm/attention/ops/triton_unified_attention.py +941 -0
  36. vllm/attention/ops/vit_attn_wrappers.py +178 -0
  37. vllm/attention/selector.py +231 -0
  38. vllm/attention/utils/__init__.py +0 -0
  39. vllm/attention/utils/fa_utils.py +109 -0
  40. vllm/attention/utils/kv_sharing_utils.py +33 -0
  41. vllm/attention/utils/kv_transfer_utils.py +60 -0
  42. vllm/beam_search.py +88 -0
  43. vllm/benchmarks/__init__.py +0 -0
  44. vllm/benchmarks/datasets.py +3222 -0
  45. vllm/benchmarks/latency.py +172 -0
  46. vllm/benchmarks/lib/__init__.py +3 -0
  47. vllm/benchmarks/lib/endpoint_request_func.py +777 -0
  48. vllm/benchmarks/lib/ready_checker.py +72 -0
  49. vllm/benchmarks/lib/utils.py +79 -0
  50. vllm/benchmarks/serve.py +1531 -0
  51. vllm/benchmarks/sweep/__init__.py +0 -0
  52. vllm/benchmarks/sweep/cli.py +38 -0
  53. vllm/benchmarks/sweep/param_sweep.py +91 -0
  54. vllm/benchmarks/sweep/plot.py +580 -0
  55. vllm/benchmarks/sweep/serve.py +416 -0
  56. vllm/benchmarks/sweep/serve_sla.py +492 -0
  57. vllm/benchmarks/sweep/server.py +114 -0
  58. vllm/benchmarks/sweep/sla_sweep.py +132 -0
  59. vllm/benchmarks/sweep/utils.py +4 -0
  60. vllm/benchmarks/throughput.py +799 -0
  61. vllm/collect_env.py +857 -0
  62. vllm/compilation/__init__.py +0 -0
  63. vllm/compilation/activation_quant_fusion.py +209 -0
  64. vllm/compilation/backends.py +759 -0
  65. vllm/compilation/base_static_graph.py +57 -0
  66. vllm/compilation/caching.py +178 -0
  67. vllm/compilation/collective_fusion.py +1234 -0
  68. vllm/compilation/compiler_interface.py +639 -0
  69. vllm/compilation/counter.py +48 -0
  70. vllm/compilation/cuda_graph.py +208 -0
  71. vllm/compilation/decorators.py +571 -0
  72. vllm/compilation/fix_functionalization.py +253 -0
  73. vllm/compilation/fusion.py +374 -0
  74. vllm/compilation/fusion_attn.py +359 -0
  75. vllm/compilation/fx_utils.py +91 -0
  76. vllm/compilation/inductor_pass.py +133 -0
  77. vllm/compilation/matcher_utils.py +317 -0
  78. vllm/compilation/monitor.py +62 -0
  79. vllm/compilation/noop_elimination.py +134 -0
  80. vllm/compilation/partition_rules.py +72 -0
  81. vllm/compilation/pass_manager.py +135 -0
  82. vllm/compilation/piecewise_backend.py +121 -0
  83. vllm/compilation/post_cleanup.py +21 -0
  84. vllm/compilation/qk_norm_rope_fusion.py +238 -0
  85. vllm/compilation/sequence_parallelism.py +363 -0
  86. vllm/compilation/torch25_custom_graph_pass.py +44 -0
  87. vllm/compilation/vllm_inductor_pass.py +173 -0
  88. vllm/compilation/wrapper.py +238 -0
  89. vllm/config/__init__.py +102 -0
  90. vllm/config/cache.py +207 -0
  91. vllm/config/compilation.py +975 -0
  92. vllm/config/device.py +75 -0
  93. vllm/config/ec_transfer.py +110 -0
  94. vllm/config/kv_events.py +56 -0
  95. vllm/config/kv_transfer.py +114 -0
  96. vllm/config/load.py +124 -0
  97. vllm/config/lora.py +112 -0
  98. vllm/config/model.py +2162 -0
  99. vllm/config/multimodal.py +248 -0
  100. vllm/config/observability.py +123 -0
  101. vllm/config/parallel.py +655 -0
  102. vllm/config/pooler.py +122 -0
  103. vllm/config/scheduler.py +298 -0
  104. vllm/config/speculative.py +654 -0
  105. vllm/config/speech_to_text.py +38 -0
  106. vllm/config/structured_outputs.py +92 -0
  107. vllm/config/utils.py +178 -0
  108. vllm/config/vllm.py +1166 -0
  109. vllm/connections.py +189 -0
  110. vllm/device_allocator/__init__.py +0 -0
  111. vllm/device_allocator/cumem.py +327 -0
  112. vllm/distributed/__init__.py +6 -0
  113. vllm/distributed/communication_op.py +43 -0
  114. vllm/distributed/device_communicators/__init__.py +0 -0
  115. vllm/distributed/device_communicators/all2all.py +490 -0
  116. vllm/distributed/device_communicators/all_reduce_utils.py +344 -0
  117. vllm/distributed/device_communicators/base_device_communicator.py +297 -0
  118. vllm/distributed/device_communicators/cpu_communicator.py +209 -0
  119. vllm/distributed/device_communicators/cuda_communicator.py +340 -0
  120. vllm/distributed/device_communicators/cuda_wrapper.py +216 -0
  121. vllm/distributed/device_communicators/custom_all_reduce.py +326 -0
  122. vllm/distributed/device_communicators/mnnvl_compat.py +27 -0
  123. vllm/distributed/device_communicators/pynccl.py +386 -0
  124. vllm/distributed/device_communicators/pynccl_allocator.py +191 -0
  125. vllm/distributed/device_communicators/pynccl_wrapper.py +564 -0
  126. vllm/distributed/device_communicators/quick_all_reduce.py +290 -0
  127. vllm/distributed/device_communicators/ray_communicator.py +259 -0
  128. vllm/distributed/device_communicators/shm_broadcast.py +733 -0
  129. vllm/distributed/device_communicators/shm_object_storage.py +660 -0
  130. vllm/distributed/device_communicators/symm_mem.py +156 -0
  131. vllm/distributed/device_communicators/tpu_communicator.py +107 -0
  132. vllm/distributed/device_communicators/xpu_communicator.py +95 -0
  133. vllm/distributed/ec_transfer/__init__.py +14 -0
  134. vllm/distributed/ec_transfer/ec_connector/__init__.py +0 -0
  135. vllm/distributed/ec_transfer/ec_connector/base.py +247 -0
  136. vllm/distributed/ec_transfer/ec_connector/factory.py +88 -0
  137. vllm/distributed/ec_transfer/ec_connector/shared_storage_connector.py +201 -0
  138. vllm/distributed/ec_transfer/ec_transfer_state.py +42 -0
  139. vllm/distributed/eplb/__init__.py +8 -0
  140. vllm/distributed/eplb/eplb_state.py +837 -0
  141. vllm/distributed/eplb/rebalance_algo.py +260 -0
  142. vllm/distributed/eplb/rebalance_execute.py +431 -0
  143. vllm/distributed/kv_events.py +371 -0
  144. vllm/distributed/kv_transfer/README.md +29 -0
  145. vllm/distributed/kv_transfer/__init__.py +20 -0
  146. vllm/distributed/kv_transfer/disagg_prefill_workflow.jpg +0 -0
  147. vllm/distributed/kv_transfer/kv_connector/__init__.py +0 -0
  148. vllm/distributed/kv_transfer/kv_connector/base.py +10 -0
  149. vllm/distributed/kv_transfer/kv_connector/factory.py +192 -0
  150. vllm/distributed/kv_transfer/kv_connector/utils.py +268 -0
  151. vllm/distributed/kv_transfer/kv_connector/v1/__init__.py +19 -0
  152. vllm/distributed/kv_transfer/kv_connector/v1/base.py +546 -0
  153. vllm/distributed/kv_transfer/kv_connector/v1/decode_bench_connector.py +419 -0
  154. vllm/distributed/kv_transfer/kv_connector/v1/lmcache_connector.py +216 -0
  155. vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/__init__.py +18 -0
  156. vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/multi_process_adapter.py +379 -0
  157. vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/utils.py +221 -0
  158. vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/vllm_v1_adapter.py +1411 -0
  159. vllm/distributed/kv_transfer/kv_connector/v1/lmcache_mp_connector.py +867 -0
  160. vllm/distributed/kv_transfer/kv_connector/v1/metrics.py +189 -0
  161. vllm/distributed/kv_transfer/kv_connector/v1/multi_connector.py +454 -0
  162. vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py +2440 -0
  163. vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py +504 -0
  164. vllm/distributed/kv_transfer/kv_connector/v1/p2p/__init__.py +0 -0
  165. vllm/distributed/kv_transfer/kv_connector/v1/p2p/p2p_nccl_connector.py +531 -0
  166. vllm/distributed/kv_transfer/kv_connector/v1/p2p/p2p_nccl_engine.py +632 -0
  167. vllm/distributed/kv_transfer/kv_connector/v1/p2p/tensor_memory_pool.py +273 -0
  168. vllm/distributed/kv_transfer/kv_connector/v1/shared_storage_connector.py +450 -0
  169. vllm/distributed/kv_transfer/kv_lookup_buffer/__init__.py +0 -0
  170. vllm/distributed/kv_transfer/kv_lookup_buffer/base.py +179 -0
  171. vllm/distributed/kv_transfer/kv_lookup_buffer/mooncake_store.py +164 -0
  172. vllm/distributed/kv_transfer/kv_lookup_buffer/simple_buffer.py +242 -0
  173. vllm/distributed/kv_transfer/kv_pipe/__init__.py +0 -0
  174. vllm/distributed/kv_transfer/kv_pipe/base.py +66 -0
  175. vllm/distributed/kv_transfer/kv_pipe/mooncake_pipe.py +295 -0
  176. vllm/distributed/kv_transfer/kv_pipe/pynccl_pipe.py +285 -0
  177. vllm/distributed/kv_transfer/kv_transfer_state.py +78 -0
  178. vllm/distributed/parallel_state.py +1759 -0
  179. vllm/distributed/tpu_distributed_utils.py +188 -0
  180. vllm/distributed/utils.py +543 -0
  181. vllm/engine/__init__.py +0 -0
  182. vllm/engine/arg_utils.py +2144 -0
  183. vllm/engine/async_llm_engine.py +6 -0
  184. vllm/engine/llm_engine.py +6 -0
  185. vllm/engine/protocol.py +170 -0
  186. vllm/entrypoints/__init__.py +0 -0
  187. vllm/entrypoints/anthropic/__init__.py +0 -0
  188. vllm/entrypoints/anthropic/protocol.py +162 -0
  189. vllm/entrypoints/anthropic/serving_messages.py +460 -0
  190. vllm/entrypoints/api_server.py +184 -0
  191. vllm/entrypoints/chat_utils.py +1690 -0
  192. vllm/entrypoints/cli/__init__.py +13 -0
  193. vllm/entrypoints/cli/benchmark/__init__.py +0 -0
  194. vllm/entrypoints/cli/benchmark/base.py +25 -0
  195. vllm/entrypoints/cli/benchmark/latency.py +21 -0
  196. vllm/entrypoints/cli/benchmark/main.py +56 -0
  197. vllm/entrypoints/cli/benchmark/serve.py +21 -0
  198. vllm/entrypoints/cli/benchmark/sweep.py +21 -0
  199. vllm/entrypoints/cli/benchmark/throughput.py +21 -0
  200. vllm/entrypoints/cli/collect_env.py +38 -0
  201. vllm/entrypoints/cli/main.py +79 -0
  202. vllm/entrypoints/cli/openai.py +256 -0
  203. vllm/entrypoints/cli/run_batch.py +68 -0
  204. vllm/entrypoints/cli/serve.py +249 -0
  205. vllm/entrypoints/cli/types.py +29 -0
  206. vllm/entrypoints/constants.py +10 -0
  207. vllm/entrypoints/context.py +572 -0
  208. vllm/entrypoints/dynamic_lora.py +57 -0
  209. vllm/entrypoints/harmony_utils.py +535 -0
  210. vllm/entrypoints/launcher.py +175 -0
  211. vllm/entrypoints/llm.py +1768 -0
  212. vllm/entrypoints/logger.py +84 -0
  213. vllm/entrypoints/openai/__init__.py +0 -0
  214. vllm/entrypoints/openai/api_server.py +2096 -0
  215. vllm/entrypoints/openai/cli_args.py +302 -0
  216. vllm/entrypoints/openai/orca_metrics.py +120 -0
  217. vllm/entrypoints/openai/protocol.py +3299 -0
  218. vllm/entrypoints/openai/run_batch.py +547 -0
  219. vllm/entrypoints/openai/serving_chat.py +1772 -0
  220. vllm/entrypoints/openai/serving_classification.py +235 -0
  221. vllm/entrypoints/openai/serving_completion.py +715 -0
  222. vllm/entrypoints/openai/serving_embedding.py +695 -0
  223. vllm/entrypoints/openai/serving_engine.py +1433 -0
  224. vllm/entrypoints/openai/serving_models.py +304 -0
  225. vllm/entrypoints/openai/serving_pooling.py +346 -0
  226. vllm/entrypoints/openai/serving_responses.py +2021 -0
  227. vllm/entrypoints/openai/serving_score.py +503 -0
  228. vllm/entrypoints/openai/serving_tokenization.py +203 -0
  229. vllm/entrypoints/openai/serving_tokens.py +269 -0
  230. vllm/entrypoints/openai/serving_transcription.py +148 -0
  231. vllm/entrypoints/openai/speech_to_text.py +405 -0
  232. vllm/entrypoints/openai/tool_parsers/__init__.py +142 -0
  233. vllm/entrypoints/openai/tool_parsers/abstract_tool_parser.py +273 -0
  234. vllm/entrypoints/openai/tool_parsers/deepseekv31_tool_parser.py +390 -0
  235. vllm/entrypoints/openai/tool_parsers/deepseekv3_tool_parser.py +390 -0
  236. vllm/entrypoints/openai/tool_parsers/ernie45_tool_parser.py +210 -0
  237. vllm/entrypoints/openai/tool_parsers/glm4_moe_tool_parser.py +200 -0
  238. vllm/entrypoints/openai/tool_parsers/granite_20b_fc_tool_parser.py +273 -0
  239. vllm/entrypoints/openai/tool_parsers/granite_tool_parser.py +253 -0
  240. vllm/entrypoints/openai/tool_parsers/hermes_tool_parser.py +494 -0
  241. vllm/entrypoints/openai/tool_parsers/hunyuan_a13b_tool_parser.py +420 -0
  242. vllm/entrypoints/openai/tool_parsers/internlm2_tool_parser.py +227 -0
  243. vllm/entrypoints/openai/tool_parsers/jamba_tool_parser.py +323 -0
  244. vllm/entrypoints/openai/tool_parsers/kimi_k2_tool_parser.py +590 -0
  245. vllm/entrypoints/openai/tool_parsers/llama4_pythonic_tool_parser.py +341 -0
  246. vllm/entrypoints/openai/tool_parsers/llama_tool_parser.py +290 -0
  247. vllm/entrypoints/openai/tool_parsers/longcat_tool_parser.py +37 -0
  248. vllm/entrypoints/openai/tool_parsers/minimax_m2_tool_parser.py +643 -0
  249. vllm/entrypoints/openai/tool_parsers/minimax_tool_parser.py +849 -0
  250. vllm/entrypoints/openai/tool_parsers/mistral_tool_parser.py +390 -0
  251. vllm/entrypoints/openai/tool_parsers/olmo3_tool_parser.py +366 -0
  252. vllm/entrypoints/openai/tool_parsers/openai_tool_parser.py +97 -0
  253. vllm/entrypoints/openai/tool_parsers/phi4mini_tool_parser.py +120 -0
  254. vllm/entrypoints/openai/tool_parsers/pythonic_tool_parser.py +332 -0
  255. vllm/entrypoints/openai/tool_parsers/qwen3coder_tool_parser.py +781 -0
  256. vllm/entrypoints/openai/tool_parsers/qwen3xml_tool_parser.py +1316 -0
  257. vllm/entrypoints/openai/tool_parsers/seed_oss_tool_parser.py +744 -0
  258. vllm/entrypoints/openai/tool_parsers/step3_tool_parser.py +303 -0
  259. vllm/entrypoints/openai/tool_parsers/utils.py +229 -0
  260. vllm/entrypoints/openai/tool_parsers/xlam_tool_parser.py +556 -0
  261. vllm/entrypoints/renderer.py +409 -0
  262. vllm/entrypoints/responses_utils.py +77 -0
  263. vllm/entrypoints/sagemaker/__init__.py +4 -0
  264. vllm/entrypoints/sagemaker/routes.py +72 -0
  265. vllm/entrypoints/score_utils.py +242 -0
  266. vllm/entrypoints/ssl.py +78 -0
  267. vllm/entrypoints/tool.py +143 -0
  268. vllm/entrypoints/tool_server.py +209 -0
  269. vllm/entrypoints/utils.py +319 -0
  270. vllm/env_override.py +378 -0
  271. vllm/envs.py +1659 -0
  272. vllm/forward_context.py +356 -0
  273. vllm/inputs/__init__.py +44 -0
  274. vllm/inputs/data.py +359 -0
  275. vllm/inputs/parse.py +137 -0
  276. vllm/inputs/preprocess.py +727 -0
  277. vllm/logger.py +267 -0
  278. vllm/logging_utils/__init__.py +10 -0
  279. vllm/logging_utils/dump_input.py +83 -0
  280. vllm/logging_utils/formatter.py +77 -0
  281. vllm/logging_utils/log_time.py +34 -0
  282. vllm/logits_process.py +121 -0
  283. vllm/logprobs.py +208 -0
  284. vllm/lora/__init__.py +0 -0
  285. vllm/lora/layers/__init__.py +41 -0
  286. vllm/lora/layers/base.py +67 -0
  287. vllm/lora/layers/base_linear.py +164 -0
  288. vllm/lora/layers/column_parallel_linear.py +578 -0
  289. vllm/lora/layers/fused_moe.py +472 -0
  290. vllm/lora/layers/logits_processor.py +252 -0
  291. vllm/lora/layers/replicated_linear.py +70 -0
  292. vllm/lora/layers/row_parallel_linear.py +181 -0
  293. vllm/lora/layers/utils.py +65 -0
  294. vllm/lora/layers/vocal_parallel_embedding.py +166 -0
  295. vllm/lora/lora_weights.py +198 -0
  296. vllm/lora/models.py +890 -0
  297. vllm/lora/ops/__init__.py +0 -0
  298. vllm/lora/ops/ipex_ops/__init__.py +6 -0
  299. vllm/lora/ops/ipex_ops/lora_ops.py +57 -0
  300. vllm/lora/ops/torch_ops/__init__.py +20 -0
  301. vllm/lora/ops/torch_ops/lora_ops.py +128 -0
  302. vllm/lora/ops/triton_ops/README_TUNING.md +60 -0
  303. vllm/lora/ops/triton_ops/__init__.py +21 -0
  304. vllm/lora/ops/triton_ops/fused_moe_lora_op.py +641 -0
  305. vllm/lora/ops/triton_ops/kernel_utils.py +340 -0
  306. vllm/lora/ops/triton_ops/lora_expand_op.py +310 -0
  307. vllm/lora/ops/triton_ops/lora_kernel_metadata.py +154 -0
  308. vllm/lora/ops/triton_ops/lora_shrink_op.py +287 -0
  309. vllm/lora/ops/triton_ops/utils.py +295 -0
  310. vllm/lora/ops/xla_ops/__init__.py +6 -0
  311. vllm/lora/ops/xla_ops/lora_ops.py +141 -0
  312. vllm/lora/peft_helper.py +128 -0
  313. vllm/lora/punica_wrapper/__init__.py +10 -0
  314. vllm/lora/punica_wrapper/punica_base.py +492 -0
  315. vllm/lora/punica_wrapper/punica_cpu.py +351 -0
  316. vllm/lora/punica_wrapper/punica_gpu.py +411 -0
  317. vllm/lora/punica_wrapper/punica_selector.py +21 -0
  318. vllm/lora/punica_wrapper/punica_tpu.py +359 -0
  319. vllm/lora/punica_wrapper/punica_xpu.py +279 -0
  320. vllm/lora/punica_wrapper/utils.py +150 -0
  321. vllm/lora/request.py +100 -0
  322. vllm/lora/resolver.py +88 -0
  323. vllm/lora/utils.py +293 -0
  324. vllm/lora/worker_manager.py +279 -0
  325. vllm/model_executor/__init__.py +11 -0
  326. vllm/model_executor/custom_op.py +194 -0
  327. vllm/model_executor/layers/__init__.py +0 -0
  328. vllm/model_executor/layers/activation.py +569 -0
  329. vllm/model_executor/layers/attention_layer_base.py +35 -0
  330. vllm/model_executor/layers/batch_invariant.py +854 -0
  331. vllm/model_executor/layers/conv.py +236 -0
  332. vllm/model_executor/layers/fla/__init__.py +8 -0
  333. vllm/model_executor/layers/fla/ops/__init__.py +17 -0
  334. vllm/model_executor/layers/fla/ops/chunk.py +240 -0
  335. vllm/model_executor/layers/fla/ops/chunk_delta_h.py +344 -0
  336. vllm/model_executor/layers/fla/ops/chunk_o.py +183 -0
  337. vllm/model_executor/layers/fla/ops/chunk_scaled_dot_kkt.py +154 -0
  338. vllm/model_executor/layers/fla/ops/cumsum.py +280 -0
  339. vllm/model_executor/layers/fla/ops/fused_recurrent.py +390 -0
  340. vllm/model_executor/layers/fla/ops/index.py +41 -0
  341. vllm/model_executor/layers/fla/ops/kda.py +1351 -0
  342. vllm/model_executor/layers/fla/ops/l2norm.py +146 -0
  343. vllm/model_executor/layers/fla/ops/layernorm_guard.py +396 -0
  344. vllm/model_executor/layers/fla/ops/op.py +60 -0
  345. vllm/model_executor/layers/fla/ops/solve_tril.py +556 -0
  346. vllm/model_executor/layers/fla/ops/utils.py +194 -0
  347. vllm/model_executor/layers/fla/ops/wy_fast.py +158 -0
  348. vllm/model_executor/layers/fused_moe/__init__.py +106 -0
  349. vllm/model_executor/layers/fused_moe/all2all_utils.py +160 -0
  350. vllm/model_executor/layers/fused_moe/batched_deep_gemm_moe.py +406 -0
  351. vllm/model_executor/layers/fused_moe/batched_triton_or_deep_gemm_moe.py +180 -0
  352. vllm/model_executor/layers/fused_moe/config.py +916 -0
  353. vllm/model_executor/layers/fused_moe/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
  354. vllm/model_executor/layers/fused_moe/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  355. vllm/model_executor/layers/fused_moe/configs/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
  356. vllm/model_executor/layers/fused_moe/configs/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
  357. vllm/model_executor/layers/fused_moe/configs/E=1,N=1792,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  358. vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
  359. vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  360. vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3.json +218 -0
  361. vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_H200,dtype=int8_w8a16.json +146 -0
  362. vllm/model_executor/layers/fused_moe/configs/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
  363. vllm/model_executor/layers/fused_moe/configs/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
  364. vllm/model_executor/layers/fused_moe/configs/E=1,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  365. vllm/model_executor/layers/fused_moe/configs/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
  366. vllm/model_executor/layers/fused_moe/configs/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
  367. vllm/model_executor/layers/fused_moe/configs/E=1,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  368. vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  369. vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=AMD_Instinct_MI300X.json +200 -0
  370. vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  371. vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=NVIDIA_H100,dtype=fp8_w8a8.json +123 -0
  372. vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  373. vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=NVIDIA_H200.json +146 -0
  374. vllm/model_executor/layers/fused_moe/configs/E=128,N=1856,device_name=NVIDIA_H100_80GB_HBM3.json +147 -0
  375. vllm/model_executor/layers/fused_moe/configs/E=128,N=1856,device_name=NVIDIA_L40S.json +147 -0
  376. vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  377. vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  378. vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H20-3e.json +146 -0
  379. vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H20.json +146 -0
  380. vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H200.json +146 -0
  381. vllm/model_executor/layers/fused_moe/configs/E=128,N=352,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +122 -0
  382. 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
  383. 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
  384. 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
  385. 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
  386. 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
  387. vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H20-3e.json +146 -0
  388. vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H20.json +146 -0
  389. 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
  390. vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H200.json +146 -0
  391. vllm/model_executor/layers/fused_moe/configs/E=128,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  392. vllm/model_executor/layers/fused_moe/configs/E=128,N=704,device_name=NVIDIA_B200,dtype=fp8_w8a8.json +146 -0
  393. vllm/model_executor/layers/fused_moe/configs/E=128,N=704,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +114 -0
  394. 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
  395. vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=AMD_Instinct_MI308X.json +213 -0
  396. 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
  397. 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
  398. 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
  399. 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
  400. vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_H20.json +146 -0
  401. 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
  402. vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_H200.json +146 -0
  403. vllm/model_executor/layers/fused_moe/configs/E=128,N=8960,device_name=NVIDIA_H100_80GB_HBM3,dtype=bf16.json +82 -0
  404. vllm/model_executor/layers/fused_moe/configs/E=128,N=8960,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +82 -0
  405. vllm/model_executor/layers/fused_moe/configs/E=128,N=928,device_name=NVIDIA_H100_80GB_HBM3.json +147 -0
  406. vllm/model_executor/layers/fused_moe/configs/E=128,N=928,device_name=NVIDIA_L40S.json +147 -0
  407. vllm/model_executor/layers/fused_moe/configs/E=128,N=96,device_name=NVIDIA_H20.json +146 -0
  408. vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=AMD_Instinct_MI300X.json +200 -0
  409. vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_B200,dtype=fp8_w8a8.json +147 -0
  410. vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_B200.json +146 -0
  411. vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_H100.json +146 -0
  412. vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  413. vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_H200.json +146 -0
  414. vllm/model_executor/layers/fused_moe/configs/E=16,N=1344,device_name=NVIDIA_A100-SXM4-40GB.json +146 -0
  415. vllm/model_executor/layers/fused_moe/configs/E=16,N=1344,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  416. vllm/model_executor/layers/fused_moe/configs/E=16,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  417. vllm/model_executor/layers/fused_moe/configs/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
  418. vllm/model_executor/layers/fused_moe/configs/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  419. vllm/model_executor/layers/fused_moe/configs/E=16,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  420. vllm/model_executor/layers/fused_moe/configs/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
  421. vllm/model_executor/layers/fused_moe/configs/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
  422. vllm/model_executor/layers/fused_moe/configs/E=16,N=1792,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  423. vllm/model_executor/layers/fused_moe/configs/E=16,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  424. vllm/model_executor/layers/fused_moe/configs/E=16,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  425. vllm/model_executor/layers/fused_moe/configs/E=16,N=2048,device_name=NVIDIA_H200.json +146 -0
  426. vllm/model_executor/layers/fused_moe/configs/E=16,N=2688,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  427. vllm/model_executor/layers/fused_moe/configs/E=16,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  428. vllm/model_executor/layers/fused_moe/configs/E=16,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
  429. vllm/model_executor/layers/fused_moe/configs/E=16,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=float8.json +146 -0
  430. vllm/model_executor/layers/fused_moe/configs/E=16,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  431. vllm/model_executor/layers/fused_moe/configs/E=16,N=3072,device_name=NVIDIA_H200,dtype=int8_w8a16.json +146 -0
  432. vllm/model_executor/layers/fused_moe/configs/E=16,N=3200,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +130 -0
  433. vllm/model_executor/layers/fused_moe/configs/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
  434. vllm/model_executor/layers/fused_moe/configs/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
  435. vllm/model_executor/layers/fused_moe/configs/E=16,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  436. vllm/model_executor/layers/fused_moe/configs/E=16,N=6400,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +130 -0
  437. vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
  438. vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  439. vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=float8.json +146 -0
  440. vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  441. vllm/model_executor/layers/fused_moe/configs/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +130 -0
  442. vllm/model_executor/layers/fused_moe/configs/E=160,N=192,device_name=AMD_Instinct_MI300X.json +201 -0
  443. vllm/model_executor/layers/fused_moe/configs/E=160,N=192,device_name=AMD_Instinct_MI350_OAM,dtype=fp8_w8a8.json +164 -0
  444. vllm/model_executor/layers/fused_moe/configs/E=160,N=192,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
  445. vllm/model_executor/layers/fused_moe/configs/E=160,N=192,device_name=NVIDIA_H20-3e.json +146 -0
  446. vllm/model_executor/layers/fused_moe/configs/E=160,N=320,device_name=NVIDIA_H20-3e.json +146 -0
  447. vllm/model_executor/layers/fused_moe/configs/E=160,N=384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  448. vllm/model_executor/layers/fused_moe/configs/E=160,N=384,device_name=AMD_Instinct_MI350_OAM,dtype=fp8_w8a8.json +164 -0
  449. vllm/model_executor/layers/fused_moe/configs/E=160,N=384,device_name=AMD_Instinct_MI355_OAM,dtype=fp8_w8a8.json +164 -0
  450. 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
  451. 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
  452. 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
  453. 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
  454. 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
  455. 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
  456. 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
  457. vllm/model_executor/layers/fused_moe/configs/E=256,N=1024,device_name=AMD_Instinct_MI325X,block_shape=[128,128].json +200 -0
  458. 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
  459. 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
  460. vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8.json +146 -0
  461. 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
  462. vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json +146 -0
  463. 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
  464. 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
  465. 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
  466. 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
  467. 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
  468. 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
  469. 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
  470. 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
  471. 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
  472. 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
  473. 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
  474. 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
  475. 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
  476. 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
  477. vllm/model_executor/layers/fused_moe/configs/E=256,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  478. vllm/model_executor/layers/fused_moe/configs/E=256,N=64,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
  479. vllm/model_executor/layers/fused_moe/configs/E=32,N=1408,device_name=NVIDIA_B200.json +147 -0
  480. 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
  481. 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
  482. 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
  483. 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
  484. 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
  485. 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
  486. 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
  487. vllm/model_executor/layers/fused_moe/configs/E=40,N=1536,device_name=NVIDIA_B200,dtype=fp8_w8a8.json +147 -0
  488. 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
  489. 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
  490. 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
  491. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_A100-SXM4-80GB.json +147 -0
  492. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_B200.json +146 -0
  493. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_GB200,dtype=fp8_w8a8.json +147 -0
  494. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  495. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_H20-3e.json +146 -0
  496. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_H200.json +146 -0
  497. vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_B200.json +146 -0
  498. vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_GB200,dtype=fp8_w8a8.json +146 -0
  499. 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
  500. vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  501. vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_H20-3e.json +146 -0
  502. vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_H200.json +146 -0
  503. vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_B200.json +146 -0
  504. vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_GB200,dtype=fp8_w8a8.json +146 -0
  505. vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  506. vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_H20-3e.json +146 -0
  507. vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_H200.json +146 -0
  508. vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_A100-SXM4-80GB.json +147 -0
  509. vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_B200.json +146 -0
  510. vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_H20-3e.json +146 -0
  511. vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_H200.json +146 -0
  512. vllm/model_executor/layers/fused_moe/configs/E=60,N=1408,device_name=AMD_Instinct_MI300X.json +200 -0
  513. vllm/model_executor/layers/fused_moe/configs/E=60,N=176,device_name=AMD_Instinct_MI300X.json +200 -0
  514. vllm/model_executor/layers/fused_moe/configs/E=60,N=352,device_name=AMD_Instinct_MI300X.json +200 -0
  515. vllm/model_executor/layers/fused_moe/configs/E=60,N=704,device_name=AMD_Instinct_MI300X.json +200 -0
  516. vllm/model_executor/layers/fused_moe/configs/E=62,N=128,device_name=AMD_Instinct_MI300X.json +200 -0
  517. vllm/model_executor/layers/fused_moe/configs/E=62,N=256,device_name=AMD_Instinct_MI300X.json +200 -0
  518. vllm/model_executor/layers/fused_moe/configs/E=62,N=256,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  519. vllm/model_executor/layers/fused_moe/configs/E=62,N=512,device_name=AMD_Instinct_MI300X.json +200 -0
  520. vllm/model_executor/layers/fused_moe/configs/E=62,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  521. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  522. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
  523. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  524. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  525. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  526. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H200.json +146 -0
  527. vllm/model_executor/layers/fused_moe/configs/E=64,N=1408,device_name=NVIDIA_B200.json +147 -0
  528. vllm/model_executor/layers/fused_moe/configs/E=64,N=1536,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
  529. vllm/model_executor/layers/fused_moe/configs/E=64,N=2560,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  530. vllm/model_executor/layers/fused_moe/configs/E=64,N=2560,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  531. vllm/model_executor/layers/fused_moe/configs/E=64,N=2560,device_name=NVIDIA_H200.json +146 -0
  532. vllm/model_executor/layers/fused_moe/configs/E=64,N=3072,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
  533. vllm/model_executor/layers/fused_moe/configs/E=64,N=3072,device_name=NVIDIA_H20.json +146 -0
  534. vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  535. vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  536. vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  537. vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H200.json +146 -0
  538. vllm/model_executor/layers/fused_moe/configs/E=64,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
  539. vllm/model_executor/layers/fused_moe/configs/E=64,N=384,device_name=NVIDIA_H20.json +146 -0
  540. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  541. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
  542. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json +146 -0
  543. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  544. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  545. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  546. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H200.json +146 -0
  547. 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
  548. vllm/model_executor/layers/fused_moe/configs/E=64,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
  549. vllm/model_executor/layers/fused_moe/configs/E=64,N=768,device_name=NVIDIA_H20.json +146 -0
  550. vllm/model_executor/layers/fused_moe/configs/E=64,N=896,device_name=NVIDIA_H20.json +146 -0
  551. vllm/model_executor/layers/fused_moe/configs/E=64,N=8960,device_name=NVIDIA_H100_80GB_HBM3,dtype=bf16.json +82 -0
  552. vllm/model_executor/layers/fused_moe/configs/E=64,N=8960,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +82 -0
  553. vllm/model_executor/layers/fused_moe/configs/E=72,N=192,device_name=AMD_Instinct_MI300X.json +200 -0
  554. vllm/model_executor/layers/fused_moe/configs/E=72,N=384,device_name=AMD_Instinct_MI300X.json +200 -0
  555. vllm/model_executor/layers/fused_moe/configs/E=72,N=384,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  556. vllm/model_executor/layers/fused_moe/configs/E=72,N=768,device_name=AMD_Instinct_MI300X.json +200 -0
  557. vllm/model_executor/layers/fused_moe/configs/E=72,N=768,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  558. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  559. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json +200 -0
  560. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  561. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI325X.json +200 -0
  562. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +138 -0
  563. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  564. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=NVIDIA_H200.json +146 -0
  565. vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  566. vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI300X.json +200 -0
  567. vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  568. vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI325X.json +200 -0
  569. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  570. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json +200 -0
  571. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  572. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI325X.json +200 -0
  573. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_A100-SXM4-40GB.json +146 -0
  574. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  575. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  576. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  577. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_H200.json +146 -0
  578. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  579. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI300X.json +200 -0
  580. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  581. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI325X.json +200 -0
  582. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  583. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  584. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  585. 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
  586. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  587. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H200.json +146 -0
  588. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  589. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json +200 -0
  590. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  591. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI325X.json +200 -0
  592. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json +146 -0
  593. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  594. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json +146 -0
  595. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  596. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  597. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  598. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H200.json +146 -0
  599. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_L40S.json +173 -0
  600. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  601. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI300X.json +200 -0
  602. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  603. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI325X.json +200 -0
  604. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  605. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  606. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  607. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  608. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H200.json +146 -0
  609. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  610. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json +200 -0
  611. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  612. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI325X.json +200 -0
  613. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  614. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  615. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  616. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  617. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H200.json +146 -0
  618. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  619. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI300X.json +200 -0
  620. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  621. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI325X.json +200 -0
  622. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  623. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  624. vllm/model_executor/layers/fused_moe/configs/README +12 -0
  625. vllm/model_executor/layers/fused_moe/cpu_fused_moe.py +354 -0
  626. vllm/model_executor/layers/fused_moe/cutlass_moe.py +1052 -0
  627. vllm/model_executor/layers/fused_moe/deep_gemm_moe.py +387 -0
  628. vllm/model_executor/layers/fused_moe/deep_gemm_utils.py +416 -0
  629. vllm/model_executor/layers/fused_moe/deepep_ht_prepare_finalize.py +420 -0
  630. vllm/model_executor/layers/fused_moe/deepep_ll_prepare_finalize.py +367 -0
  631. vllm/model_executor/layers/fused_moe/flashinfer_cutlass_moe.py +307 -0
  632. vllm/model_executor/layers/fused_moe/flashinfer_cutlass_prepare_finalize.py +362 -0
  633. vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py +192 -0
  634. vllm/model_executor/layers/fused_moe/fused_batched_moe.py +1012 -0
  635. vllm/model_executor/layers/fused_moe/fused_marlin_moe.py +792 -0
  636. vllm/model_executor/layers/fused_moe/fused_moe.py +2175 -0
  637. vllm/model_executor/layers/fused_moe/fused_moe_method_base.py +112 -0
  638. vllm/model_executor/layers/fused_moe/fused_moe_modular_method.py +164 -0
  639. vllm/model_executor/layers/fused_moe/gpt_oss_triton_kernels_moe.py +316 -0
  640. vllm/model_executor/layers/fused_moe/layer.py +1944 -0
  641. vllm/model_executor/layers/fused_moe/modular_kernel.py +1222 -0
  642. vllm/model_executor/layers/fused_moe/moe_align_block_size.py +174 -0
  643. vllm/model_executor/layers/fused_moe/moe_pallas.py +83 -0
  644. vllm/model_executor/layers/fused_moe/moe_permute_unpermute.py +229 -0
  645. vllm/model_executor/layers/fused_moe/moe_torch_iterative.py +60 -0
  646. vllm/model_executor/layers/fused_moe/pplx_prepare_finalize.py +362 -0
  647. vllm/model_executor/layers/fused_moe/prepare_finalize.py +77 -0
  648. vllm/model_executor/layers/fused_moe/rocm_aiter_fused_moe.py +265 -0
  649. vllm/model_executor/layers/fused_moe/routing_simulator.py +310 -0
  650. vllm/model_executor/layers/fused_moe/shared_fused_moe.py +97 -0
  651. vllm/model_executor/layers/fused_moe/topk_weight_and_reduce.py +171 -0
  652. vllm/model_executor/layers/fused_moe/triton_deep_gemm_moe.py +163 -0
  653. vllm/model_executor/layers/fused_moe/trtllm_moe.py +143 -0
  654. vllm/model_executor/layers/fused_moe/unquantized_fused_moe_method.py +578 -0
  655. vllm/model_executor/layers/fused_moe/utils.py +332 -0
  656. vllm/model_executor/layers/kda.py +448 -0
  657. vllm/model_executor/layers/layernorm.py +442 -0
  658. vllm/model_executor/layers/lightning_attn.py +729 -0
  659. vllm/model_executor/layers/linear.py +1424 -0
  660. vllm/model_executor/layers/logits_processor.py +106 -0
  661. vllm/model_executor/layers/mamba/__init__.py +0 -0
  662. vllm/model_executor/layers/mamba/abstract.py +71 -0
  663. vllm/model_executor/layers/mamba/linear_attn.py +402 -0
  664. vllm/model_executor/layers/mamba/mamba_mixer.py +535 -0
  665. vllm/model_executor/layers/mamba/mamba_mixer2.py +928 -0
  666. vllm/model_executor/layers/mamba/mamba_utils.py +225 -0
  667. vllm/model_executor/layers/mamba/ops/__init__.py +0 -0
  668. vllm/model_executor/layers/mamba/ops/causal_conv1d.py +1240 -0
  669. vllm/model_executor/layers/mamba/ops/layernorm_gated.py +172 -0
  670. vllm/model_executor/layers/mamba/ops/mamba_ssm.py +478 -0
  671. vllm/model_executor/layers/mamba/ops/ssd_bmm.py +211 -0
  672. vllm/model_executor/layers/mamba/ops/ssd_chunk_scan.py +456 -0
  673. vllm/model_executor/layers/mamba/ops/ssd_chunk_state.py +700 -0
  674. vllm/model_executor/layers/mamba/ops/ssd_combined.py +230 -0
  675. vllm/model_executor/layers/mamba/ops/ssd_state_passing.py +157 -0
  676. vllm/model_executor/layers/mamba/short_conv.py +264 -0
  677. vllm/model_executor/layers/mla.py +168 -0
  678. vllm/model_executor/layers/pooler.py +817 -0
  679. vllm/model_executor/layers/quantization/__init__.py +174 -0
  680. vllm/model_executor/layers/quantization/auto_round.py +454 -0
  681. vllm/model_executor/layers/quantization/awq.py +277 -0
  682. vllm/model_executor/layers/quantization/awq_marlin.py +659 -0
  683. vllm/model_executor/layers/quantization/awq_triton.py +337 -0
  684. vllm/model_executor/layers/quantization/base_config.py +170 -0
  685. vllm/model_executor/layers/quantization/bitblas.py +502 -0
  686. vllm/model_executor/layers/quantization/bitsandbytes.py +658 -0
  687. vllm/model_executor/layers/quantization/compressed_tensors/__init__.py +3 -0
  688. vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py +914 -0
  689. vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py +2284 -0
  690. vllm/model_executor/layers/quantization/compressed_tensors/schemes/__init__.py +35 -0
  691. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_24.py +392 -0
  692. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_scheme.py +55 -0
  693. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a16_24.py +176 -0
  694. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a16_nvfp4.py +124 -0
  695. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_nvfp4.py +218 -0
  696. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a8_fp8.py +183 -0
  697. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a8_int.py +153 -0
  698. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a16_fp8.py +138 -0
  699. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py +200 -0
  700. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py +125 -0
  701. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16.py +219 -0
  702. vllm/model_executor/layers/quantization/compressed_tensors/transform/__init__.py +0 -0
  703. vllm/model_executor/layers/quantization/compressed_tensors/transform/linear.py +260 -0
  704. vllm/model_executor/layers/quantization/compressed_tensors/transform/module.py +173 -0
  705. vllm/model_executor/layers/quantization/compressed_tensors/transform/schemes/__init__.py +0 -0
  706. vllm/model_executor/layers/quantization/compressed_tensors/transform/schemes/linear_qutlass_nvfp4.py +64 -0
  707. vllm/model_executor/layers/quantization/compressed_tensors/transform/utils.py +13 -0
  708. vllm/model_executor/layers/quantization/compressed_tensors/triton_scaled_mm.py +224 -0
  709. vllm/model_executor/layers/quantization/compressed_tensors/utils.py +216 -0
  710. vllm/model_executor/layers/quantization/deepspeedfp.py +218 -0
  711. vllm/model_executor/layers/quantization/experts_int8.py +240 -0
  712. vllm/model_executor/layers/quantization/fbgemm_fp8.py +195 -0
  713. vllm/model_executor/layers/quantization/fp8.py +1333 -0
  714. vllm/model_executor/layers/quantization/fp_quant.py +420 -0
  715. vllm/model_executor/layers/quantization/gguf.py +643 -0
  716. vllm/model_executor/layers/quantization/gptq.py +393 -0
  717. vllm/model_executor/layers/quantization/gptq_bitblas.py +482 -0
  718. vllm/model_executor/layers/quantization/gptq_marlin.py +789 -0
  719. vllm/model_executor/layers/quantization/gptq_marlin_24.py +320 -0
  720. vllm/model_executor/layers/quantization/hqq_marlin.py +371 -0
  721. vllm/model_executor/layers/quantization/inc.py +65 -0
  722. vllm/model_executor/layers/quantization/input_quant_fp8.py +171 -0
  723. vllm/model_executor/layers/quantization/ipex_quant.py +467 -0
  724. vllm/model_executor/layers/quantization/kernels/__init__.py +0 -0
  725. vllm/model_executor/layers/quantization/kernels/mixed_precision/MPLinearKernel.py +94 -0
  726. vllm/model_executor/layers/quantization/kernels/mixed_precision/__init__.py +105 -0
  727. vllm/model_executor/layers/quantization/kernels/mixed_precision/allspark.py +115 -0
  728. vllm/model_executor/layers/quantization/kernels/mixed_precision/bitblas.py +323 -0
  729. vllm/model_executor/layers/quantization/kernels/mixed_precision/conch.py +98 -0
  730. vllm/model_executor/layers/quantization/kernels/mixed_precision/cutlass.py +119 -0
  731. vllm/model_executor/layers/quantization/kernels/mixed_precision/dynamic_4bit.py +111 -0
  732. vllm/model_executor/layers/quantization/kernels/mixed_precision/exllama.py +161 -0
  733. vllm/model_executor/layers/quantization/kernels/mixed_precision/machete.py +159 -0
  734. vllm/model_executor/layers/quantization/kernels/mixed_precision/marlin.py +166 -0
  735. vllm/model_executor/layers/quantization/kernels/scaled_mm/ScaledMMLinearKernel.py +73 -0
  736. vllm/model_executor/layers/quantization/kernels/scaled_mm/__init__.py +97 -0
  737. vllm/model_executor/layers/quantization/kernels/scaled_mm/aiter.py +120 -0
  738. vllm/model_executor/layers/quantization/kernels/scaled_mm/cpu.py +219 -0
  739. vllm/model_executor/layers/quantization/kernels/scaled_mm/cutlass.py +140 -0
  740. vllm/model_executor/layers/quantization/kernels/scaled_mm/triton.py +42 -0
  741. vllm/model_executor/layers/quantization/kernels/scaled_mm/xla.py +105 -0
  742. vllm/model_executor/layers/quantization/kv_cache.py +146 -0
  743. vllm/model_executor/layers/quantization/modelopt.py +1788 -0
  744. vllm/model_executor/layers/quantization/moe_wna16.py +541 -0
  745. vllm/model_executor/layers/quantization/mxfp4.py +1162 -0
  746. vllm/model_executor/layers/quantization/petit.py +320 -0
  747. vllm/model_executor/layers/quantization/ptpc_fp8.py +137 -0
  748. vllm/model_executor/layers/quantization/quark/__init__.py +0 -0
  749. vllm/model_executor/layers/quantization/quark/quark.py +528 -0
  750. vllm/model_executor/layers/quantization/quark/quark_moe.py +683 -0
  751. vllm/model_executor/layers/quantization/quark/schemes/__init__.py +9 -0
  752. vllm/model_executor/layers/quantization/quark/schemes/quark_ocp_mx.py +306 -0
  753. vllm/model_executor/layers/quantization/quark/schemes/quark_scheme.py +55 -0
  754. vllm/model_executor/layers/quantization/quark/schemes/quark_w8a8_fp8.py +179 -0
  755. vllm/model_executor/layers/quantization/quark/schemes/quark_w8a8_int8.py +139 -0
  756. vllm/model_executor/layers/quantization/quark/utils.py +105 -0
  757. vllm/model_executor/layers/quantization/qutlass_utils.py +185 -0
  758. vllm/model_executor/layers/quantization/rtn.py +652 -0
  759. vllm/model_executor/layers/quantization/schema.py +90 -0
  760. vllm/model_executor/layers/quantization/torchao.py +380 -0
  761. vllm/model_executor/layers/quantization/tpu_int8.py +139 -0
  762. vllm/model_executor/layers/quantization/utils/__init__.py +6 -0
  763. vllm/model_executor/layers/quantization/utils/allspark_utils.py +67 -0
  764. vllm/model_executor/layers/quantization/utils/bitblas_utils.py +229 -0
  765. 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
  766. 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
  767. 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
  768. 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
  769. 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
  770. 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
  771. 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
  772. 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
  773. 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
  774. 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
  775. 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
  776. 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
  777. 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
  778. 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
  779. 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
  780. 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
  781. 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
  782. 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
  783. 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
  784. 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
  785. 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
  786. 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
  787. 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
  788. 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
  789. 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
  790. 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
  791. 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
  792. 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
  793. 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
  794. 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
  795. 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
  796. 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
  797. 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
  798. 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
  799. 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
  800. 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
  801. 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
  802. 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
  803. 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
  804. 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
  805. 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
  806. 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
  807. 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
  808. 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
  809. 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
  810. 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
  811. 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
  812. 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
  813. 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
  814. 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
  815. 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
  816. 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
  817. 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
  818. 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
  819. 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
  820. 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
  821. 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
  822. 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
  823. 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
  824. 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
  825. 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
  826. 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
  827. 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
  828. 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
  829. 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
  830. 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
  831. 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
  832. 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
  833. 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
  834. 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
  835. 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
  836. 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
  837. 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
  838. 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
  839. 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
  840. 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
  841. 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
  842. 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
  843. 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
  844. 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
  845. 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
  846. 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
  847. 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
  848. 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
  849. 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
  850. 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
  851. 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
  852. 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
  853. 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
  854. 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
  855. 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
  856. 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
  857. 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
  858. 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
  859. 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
  860. 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
  861. 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
  862. 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
  863. 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
  864. 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
  865. 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
  866. 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
  867. 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
  868. 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
  869. 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
  870. 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
  871. 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
  872. 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
  873. 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
  874. 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
  875. 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
  876. 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
  877. 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
  878. 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
  879. 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
  880. 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
  881. 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
  882. 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
  883. 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
  884. 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
  885. 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
  886. 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
  887. 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
  888. 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
  889. 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
  890. 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
  891. 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
  892. 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
  893. 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
  894. 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
  895. 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
  896. 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
  897. 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
  898. 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
  899. 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
  900. 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
  901. 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
  902. 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
  903. 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
  904. 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
  905. 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
  906. 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
  907. 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
  908. 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
  909. 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
  910. 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
  911. 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
  912. 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
  913. 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
  914. 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
  915. 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
  916. 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
  917. 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
  918. 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
  919. 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
  920. 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
  921. 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
  922. 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
  923. 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
  924. 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
  925. 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
  926. 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
  927. 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
  928. 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
  929. 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
  930. 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
  931. 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
  932. 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
  933. 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
  934. 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
  935. 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
  936. 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
  937. 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
  938. 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
  939. 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
  940. 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
  941. 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
  942. 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
  943. 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
  944. 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
  945. 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
  946. 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
  947. 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
  948. 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
  949. 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
  950. 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
  951. 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
  952. 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
  953. 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
  954. 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
  955. 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
  956. 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
  957. 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
  958. 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
  959. 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
  960. 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
  961. 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
  962. 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
  963. 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
  964. 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
  965. 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
  966. 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
  967. 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
  968. 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
  969. 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
  970. 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
  971. 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
  972. 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
  973. 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
  974. 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
  975. vllm/model_executor/layers/quantization/utils/configs/README.md +3 -0
  976. vllm/model_executor/layers/quantization/utils/flashinfer_fp4_moe.py +89 -0
  977. vllm/model_executor/layers/quantization/utils/flashinfer_utils.py +298 -0
  978. vllm/model_executor/layers/quantization/utils/fp8_utils.py +1203 -0
  979. vllm/model_executor/layers/quantization/utils/gptq_utils.py +158 -0
  980. vllm/model_executor/layers/quantization/utils/int8_utils.py +489 -0
  981. vllm/model_executor/layers/quantization/utils/layer_utils.py +41 -0
  982. vllm/model_executor/layers/quantization/utils/machete_utils.py +56 -0
  983. vllm/model_executor/layers/quantization/utils/marlin_utils.py +575 -0
  984. vllm/model_executor/layers/quantization/utils/marlin_utils_fp4.py +397 -0
  985. vllm/model_executor/layers/quantization/utils/marlin_utils_fp8.py +351 -0
  986. vllm/model_executor/layers/quantization/utils/marlin_utils_test.py +161 -0
  987. vllm/model_executor/layers/quantization/utils/marlin_utils_test_24.py +467 -0
  988. vllm/model_executor/layers/quantization/utils/mxfp4_utils.py +181 -0
  989. vllm/model_executor/layers/quantization/utils/mxfp6_utils.py +142 -0
  990. vllm/model_executor/layers/quantization/utils/mxfp8_utils.py +24 -0
  991. vllm/model_executor/layers/quantization/utils/nvfp4_emulation_utils.py +142 -0
  992. vllm/model_executor/layers/quantization/utils/nvfp4_moe_support.py +63 -0
  993. vllm/model_executor/layers/quantization/utils/ocp_mx_utils.py +51 -0
  994. vllm/model_executor/layers/quantization/utils/petit_utils.py +124 -0
  995. vllm/model_executor/layers/quantization/utils/quant_utils.py +687 -0
  996. vllm/model_executor/layers/quantization/utils/w8a8_utils.py +516 -0
  997. vllm/model_executor/layers/resampler.py +283 -0
  998. vllm/model_executor/layers/rotary_embedding/__init__.py +278 -0
  999. vllm/model_executor/layers/rotary_embedding/base.py +235 -0
  1000. vllm/model_executor/layers/rotary_embedding/common.py +188 -0
  1001. vllm/model_executor/layers/rotary_embedding/deepseek_scaling_rope.py +165 -0
  1002. vllm/model_executor/layers/rotary_embedding/dual_chunk_rope.py +215 -0
  1003. vllm/model_executor/layers/rotary_embedding/dynamic_ntk_alpha_rope.py +43 -0
  1004. vllm/model_executor/layers/rotary_embedding/dynamic_ntk_scaling_rope.py +68 -0
  1005. vllm/model_executor/layers/rotary_embedding/ernie45_vl_rope.py +75 -0
  1006. vllm/model_executor/layers/rotary_embedding/linear_scaling_rope.py +115 -0
  1007. vllm/model_executor/layers/rotary_embedding/llama3_rope.py +54 -0
  1008. vllm/model_executor/layers/rotary_embedding/llama4_vision_rope.py +80 -0
  1009. vllm/model_executor/layers/rotary_embedding/mrope.py +397 -0
  1010. vllm/model_executor/layers/rotary_embedding/ntk_scaling_rope.py +47 -0
  1011. vllm/model_executor/layers/rotary_embedding/phi3_long_rope_scaled_rope.py +159 -0
  1012. vllm/model_executor/layers/rotary_embedding/yarn_scaling_rope.py +81 -0
  1013. vllm/model_executor/layers/utils.py +251 -0
  1014. vllm/model_executor/layers/vocab_parallel_embedding.py +558 -0
  1015. vllm/model_executor/model_loader/__init__.py +148 -0
  1016. vllm/model_executor/model_loader/base_loader.py +57 -0
  1017. vllm/model_executor/model_loader/bitsandbytes_loader.py +822 -0
  1018. vllm/model_executor/model_loader/default_loader.py +327 -0
  1019. vllm/model_executor/model_loader/dummy_loader.py +28 -0
  1020. vllm/model_executor/model_loader/gguf_loader.py +176 -0
  1021. vllm/model_executor/model_loader/online_quantization.py +224 -0
  1022. vllm/model_executor/model_loader/runai_streamer_loader.py +116 -0
  1023. vllm/model_executor/model_loader/sharded_state_loader.py +206 -0
  1024. vllm/model_executor/model_loader/tensorizer.py +790 -0
  1025. vllm/model_executor/model_loader/tensorizer_loader.py +151 -0
  1026. vllm/model_executor/model_loader/tpu.py +118 -0
  1027. vllm/model_executor/model_loader/utils.py +288 -0
  1028. vllm/model_executor/model_loader/weight_utils.py +1084 -0
  1029. vllm/model_executor/models/__init__.py +44 -0
  1030. vllm/model_executor/models/adapters.py +543 -0
  1031. vllm/model_executor/models/afmoe.py +711 -0
  1032. vllm/model_executor/models/aimv2.py +247 -0
  1033. vllm/model_executor/models/apertus.py +587 -0
  1034. vllm/model_executor/models/arcee.py +439 -0
  1035. vllm/model_executor/models/arctic.py +635 -0
  1036. vllm/model_executor/models/aria.py +655 -0
  1037. vllm/model_executor/models/aya_vision.py +450 -0
  1038. vllm/model_executor/models/baichuan.py +496 -0
  1039. vllm/model_executor/models/bailing_moe.py +646 -0
  1040. vllm/model_executor/models/bamba.py +522 -0
  1041. vllm/model_executor/models/bee.py +157 -0
  1042. vllm/model_executor/models/bert.py +925 -0
  1043. vllm/model_executor/models/bert_with_rope.py +732 -0
  1044. vllm/model_executor/models/blip.py +349 -0
  1045. vllm/model_executor/models/blip2.py +695 -0
  1046. vllm/model_executor/models/bloom.py +390 -0
  1047. vllm/model_executor/models/chameleon.py +1120 -0
  1048. vllm/model_executor/models/chatglm.py +498 -0
  1049. vllm/model_executor/models/clip.py +965 -0
  1050. vllm/model_executor/models/cohere2_vision.py +472 -0
  1051. vllm/model_executor/models/commandr.py +473 -0
  1052. vllm/model_executor/models/config.py +503 -0
  1053. vllm/model_executor/models/dbrx.py +482 -0
  1054. vllm/model_executor/models/deepencoder.py +673 -0
  1055. vllm/model_executor/models/deepseek_eagle.py +260 -0
  1056. vllm/model_executor/models/deepseek_mtp.py +360 -0
  1057. vllm/model_executor/models/deepseek_ocr.py +593 -0
  1058. vllm/model_executor/models/deepseek_v2.py +1649 -0
  1059. vllm/model_executor/models/deepseek_vl2.py +655 -0
  1060. vllm/model_executor/models/dots1.py +574 -0
  1061. vllm/model_executor/models/dots_ocr.py +900 -0
  1062. vllm/model_executor/models/ernie45.py +53 -0
  1063. vllm/model_executor/models/ernie45_moe.py +759 -0
  1064. vllm/model_executor/models/ernie45_vl.py +1742 -0
  1065. vllm/model_executor/models/ernie45_vl_moe.py +803 -0
  1066. vllm/model_executor/models/ernie_mtp.py +279 -0
  1067. vllm/model_executor/models/exaone.py +545 -0
  1068. vllm/model_executor/models/exaone4.py +531 -0
  1069. vllm/model_executor/models/fairseq2_llama.py +154 -0
  1070. vllm/model_executor/models/falcon.py +545 -0
  1071. vllm/model_executor/models/falcon_h1.py +685 -0
  1072. vllm/model_executor/models/flex_olmo.py +155 -0
  1073. vllm/model_executor/models/fuyu.py +373 -0
  1074. vllm/model_executor/models/gemma.py +426 -0
  1075. vllm/model_executor/models/gemma2.py +439 -0
  1076. vllm/model_executor/models/gemma3.py +571 -0
  1077. vllm/model_executor/models/gemma3_mm.py +741 -0
  1078. vllm/model_executor/models/gemma3n.py +1165 -0
  1079. vllm/model_executor/models/gemma3n_mm.py +811 -0
  1080. vllm/model_executor/models/glm.py +23 -0
  1081. vllm/model_executor/models/glm4.py +305 -0
  1082. vllm/model_executor/models/glm4_1v.py +1821 -0
  1083. vllm/model_executor/models/glm4_moe.py +747 -0
  1084. vllm/model_executor/models/glm4_moe_mtp.py +359 -0
  1085. vllm/model_executor/models/glm4v.py +784 -0
  1086. vllm/model_executor/models/gpt2.py +397 -0
  1087. vllm/model_executor/models/gpt_bigcode.py +339 -0
  1088. vllm/model_executor/models/gpt_j.py +346 -0
  1089. vllm/model_executor/models/gpt_neox.py +344 -0
  1090. vllm/model_executor/models/gpt_oss.py +738 -0
  1091. vllm/model_executor/models/granite.py +516 -0
  1092. vllm/model_executor/models/granite_speech.py +913 -0
  1093. vllm/model_executor/models/granitemoe.py +569 -0
  1094. vllm/model_executor/models/granitemoehybrid.py +709 -0
  1095. vllm/model_executor/models/granitemoeshared.py +333 -0
  1096. vllm/model_executor/models/gritlm.py +245 -0
  1097. vllm/model_executor/models/grok1.py +558 -0
  1098. vllm/model_executor/models/h2ovl.py +554 -0
  1099. vllm/model_executor/models/hunyuan_v1.py +1053 -0
  1100. vllm/model_executor/models/hyperclovax_vision.py +1166 -0
  1101. vllm/model_executor/models/idefics2_vision_model.py +426 -0
  1102. vllm/model_executor/models/idefics3.py +717 -0
  1103. vllm/model_executor/models/interfaces.py +1092 -0
  1104. vllm/model_executor/models/interfaces_base.py +214 -0
  1105. vllm/model_executor/models/intern_vit.py +453 -0
  1106. vllm/model_executor/models/internlm2.py +460 -0
  1107. vllm/model_executor/models/internlm2_ve.py +142 -0
  1108. vllm/model_executor/models/interns1.py +830 -0
  1109. vllm/model_executor/models/interns1_vit.py +432 -0
  1110. vllm/model_executor/models/internvl.py +1452 -0
  1111. vllm/model_executor/models/jais.py +397 -0
  1112. vllm/model_executor/models/jamba.py +610 -0
  1113. vllm/model_executor/models/jina_vl.py +147 -0
  1114. vllm/model_executor/models/keye.py +1761 -0
  1115. vllm/model_executor/models/keye_vl1_5.py +726 -0
  1116. vllm/model_executor/models/kimi_linear.py +663 -0
  1117. vllm/model_executor/models/kimi_vl.py +578 -0
  1118. vllm/model_executor/models/lfm2.py +532 -0
  1119. vllm/model_executor/models/lfm2_moe.py +762 -0
  1120. vllm/model_executor/models/lightonocr.py +195 -0
  1121. vllm/model_executor/models/llama.py +732 -0
  1122. vllm/model_executor/models/llama4.py +859 -0
  1123. vllm/model_executor/models/llama4_eagle.py +223 -0
  1124. vllm/model_executor/models/llama_eagle.py +218 -0
  1125. vllm/model_executor/models/llama_eagle3.py +367 -0
  1126. vllm/model_executor/models/llava.py +842 -0
  1127. vllm/model_executor/models/llava_next.py +583 -0
  1128. vllm/model_executor/models/llava_next_video.py +467 -0
  1129. vllm/model_executor/models/llava_onevision.py +923 -0
  1130. vllm/model_executor/models/longcat_flash.py +749 -0
  1131. vllm/model_executor/models/longcat_flash_mtp.py +349 -0
  1132. vllm/model_executor/models/mamba.py +276 -0
  1133. vllm/model_executor/models/mamba2.py +289 -0
  1134. vllm/model_executor/models/medusa.py +179 -0
  1135. vllm/model_executor/models/midashenglm.py +827 -0
  1136. vllm/model_executor/models/mimo.py +188 -0
  1137. vllm/model_executor/models/mimo_mtp.py +294 -0
  1138. vllm/model_executor/models/minicpm.py +664 -0
  1139. vllm/model_executor/models/minicpm3.py +242 -0
  1140. vllm/model_executor/models/minicpm_eagle.py +389 -0
  1141. vllm/model_executor/models/minicpmo.py +768 -0
  1142. vllm/model_executor/models/minicpmv.py +1745 -0
  1143. vllm/model_executor/models/minimax_m2.py +552 -0
  1144. vllm/model_executor/models/minimax_text_01.py +1012 -0
  1145. vllm/model_executor/models/minimax_vl_01.py +396 -0
  1146. vllm/model_executor/models/mistral3.py +637 -0
  1147. vllm/model_executor/models/mixtral.py +621 -0
  1148. vllm/model_executor/models/mllama4.py +1147 -0
  1149. vllm/model_executor/models/mlp_speculator.py +235 -0
  1150. vllm/model_executor/models/modernbert.py +450 -0
  1151. vllm/model_executor/models/module_mapping.py +74 -0
  1152. vllm/model_executor/models/molmo.py +1555 -0
  1153. vllm/model_executor/models/moonvit.py +677 -0
  1154. vllm/model_executor/models/mpt.py +335 -0
  1155. vllm/model_executor/models/nano_nemotron_vl.py +1740 -0
  1156. vllm/model_executor/models/nemotron.py +518 -0
  1157. vllm/model_executor/models/nemotron_h.py +852 -0
  1158. vllm/model_executor/models/nemotron_nas.py +491 -0
  1159. vllm/model_executor/models/nemotron_vl.py +653 -0
  1160. vllm/model_executor/models/nvlm_d.py +216 -0
  1161. vllm/model_executor/models/olmo.py +414 -0
  1162. vllm/model_executor/models/olmo2.py +454 -0
  1163. vllm/model_executor/models/olmoe.py +498 -0
  1164. vllm/model_executor/models/openpangu.py +1062 -0
  1165. vllm/model_executor/models/openpangu_mtp.py +265 -0
  1166. vllm/model_executor/models/opt.py +426 -0
  1167. vllm/model_executor/models/orion.py +372 -0
  1168. vllm/model_executor/models/ouro.py +516 -0
  1169. vllm/model_executor/models/ovis.py +559 -0
  1170. vllm/model_executor/models/ovis2_5.py +673 -0
  1171. vllm/model_executor/models/paddleocr_vl.py +1407 -0
  1172. vllm/model_executor/models/paligemma.py +412 -0
  1173. vllm/model_executor/models/persimmon.py +377 -0
  1174. vllm/model_executor/models/phi.py +374 -0
  1175. vllm/model_executor/models/phi3.py +18 -0
  1176. vllm/model_executor/models/phi3v.py +737 -0
  1177. vllm/model_executor/models/phi4_multimodal.py +1447 -0
  1178. vllm/model_executor/models/phi4mm.py +1253 -0
  1179. vllm/model_executor/models/phi4mm_audio.py +1296 -0
  1180. vllm/model_executor/models/phi4mm_utils.py +1907 -0
  1181. vllm/model_executor/models/phimoe.py +675 -0
  1182. vllm/model_executor/models/pixtral.py +1352 -0
  1183. vllm/model_executor/models/plamo2.py +981 -0
  1184. vllm/model_executor/models/qwen.py +368 -0
  1185. vllm/model_executor/models/qwen2.py +541 -0
  1186. vllm/model_executor/models/qwen2_5_omni_thinker.py +1246 -0
  1187. vllm/model_executor/models/qwen2_5_vl.py +1613 -0
  1188. vllm/model_executor/models/qwen2_audio.py +473 -0
  1189. vllm/model_executor/models/qwen2_moe.py +596 -0
  1190. vllm/model_executor/models/qwen2_rm.py +123 -0
  1191. vllm/model_executor/models/qwen2_vl.py +1670 -0
  1192. vllm/model_executor/models/qwen3.py +336 -0
  1193. vllm/model_executor/models/qwen3_moe.py +744 -0
  1194. vllm/model_executor/models/qwen3_next.py +1395 -0
  1195. vllm/model_executor/models/qwen3_next_mtp.py +296 -0
  1196. vllm/model_executor/models/qwen3_omni_moe_thinker.py +1721 -0
  1197. vllm/model_executor/models/qwen3_vl.py +1673 -0
  1198. vllm/model_executor/models/qwen3_vl_moe.py +415 -0
  1199. vllm/model_executor/models/qwen_vl.py +802 -0
  1200. vllm/model_executor/models/radio.py +555 -0
  1201. vllm/model_executor/models/registry.py +1155 -0
  1202. vllm/model_executor/models/roberta.py +259 -0
  1203. vllm/model_executor/models/rvl.py +107 -0
  1204. vllm/model_executor/models/seed_oss.py +497 -0
  1205. vllm/model_executor/models/siglip.py +1174 -0
  1206. vllm/model_executor/models/siglip2navit.py +724 -0
  1207. vllm/model_executor/models/skyworkr1v.py +953 -0
  1208. vllm/model_executor/models/smolvlm.py +38 -0
  1209. vllm/model_executor/models/solar.py +502 -0
  1210. vllm/model_executor/models/stablelm.py +359 -0
  1211. vllm/model_executor/models/starcoder2.py +367 -0
  1212. vllm/model_executor/models/step3_text.py +559 -0
  1213. vllm/model_executor/models/step3_vl.py +1148 -0
  1214. vllm/model_executor/models/swin.py +514 -0
  1215. vllm/model_executor/models/tarsier.py +619 -0
  1216. vllm/model_executor/models/telechat2.py +153 -0
  1217. vllm/model_executor/models/teleflm.py +78 -0
  1218. vllm/model_executor/models/terratorch.py +319 -0
  1219. vllm/model_executor/models/transformers/__init__.py +127 -0
  1220. vllm/model_executor/models/transformers/base.py +464 -0
  1221. vllm/model_executor/models/transformers/causal.py +65 -0
  1222. vllm/model_executor/models/transformers/legacy.py +90 -0
  1223. vllm/model_executor/models/transformers/moe.py +318 -0
  1224. vllm/model_executor/models/transformers/multimodal.py +411 -0
  1225. vllm/model_executor/models/transformers/pooling.py +119 -0
  1226. vllm/model_executor/models/transformers/utils.py +207 -0
  1227. vllm/model_executor/models/ultravox.py +681 -0
  1228. vllm/model_executor/models/utils.py +877 -0
  1229. vllm/model_executor/models/vision.py +552 -0
  1230. vllm/model_executor/models/voxtral.py +845 -0
  1231. vllm/model_executor/models/whisper.py +959 -0
  1232. vllm/model_executor/models/zamba2.py +986 -0
  1233. vllm/model_executor/parameter.py +642 -0
  1234. vllm/model_executor/utils.py +94 -0
  1235. vllm/model_executor/warmup/__init__.py +0 -0
  1236. vllm/model_executor/warmup/deep_gemm_warmup.py +314 -0
  1237. vllm/model_executor/warmup/kernel_warmup.py +98 -0
  1238. vllm/multimodal/__init__.py +40 -0
  1239. vllm/multimodal/audio.py +118 -0
  1240. vllm/multimodal/base.py +26 -0
  1241. vllm/multimodal/cache.py +755 -0
  1242. vllm/multimodal/evs.py +294 -0
  1243. vllm/multimodal/hasher.py +106 -0
  1244. vllm/multimodal/image.py +130 -0
  1245. vllm/multimodal/inputs.py +1036 -0
  1246. vllm/multimodal/parse.py +544 -0
  1247. vllm/multimodal/processing.py +2186 -0
  1248. vllm/multimodal/profiling.py +369 -0
  1249. vllm/multimodal/registry.py +360 -0
  1250. vllm/multimodal/utils.py +512 -0
  1251. vllm/multimodal/video.py +306 -0
  1252. vllm/outputs.py +345 -0
  1253. vllm/platforms/__init__.py +277 -0
  1254. vllm/platforms/cpu.py +414 -0
  1255. vllm/platforms/cuda.py +657 -0
  1256. vllm/platforms/interface.py +639 -0
  1257. vllm/platforms/rocm.py +466 -0
  1258. vllm/platforms/tpu.py +276 -0
  1259. vllm/platforms/xpu.py +274 -0
  1260. vllm/plugins/__init__.py +78 -0
  1261. vllm/plugins/io_processors/__init__.py +68 -0
  1262. vllm/plugins/io_processors/interface.py +77 -0
  1263. vllm/plugins/lora_resolvers/__init__.py +0 -0
  1264. vllm/plugins/lora_resolvers/filesystem_resolver.py +52 -0
  1265. vllm/pooling_params.py +228 -0
  1266. vllm/profiler/__init__.py +0 -0
  1267. vllm/profiler/gpu_profiler.py +37 -0
  1268. vllm/profiler/layerwise_profile.py +392 -0
  1269. vllm/profiler/utils.py +151 -0
  1270. vllm/py.typed +2 -0
  1271. vllm/ray/__init__.py +0 -0
  1272. vllm/ray/lazy_utils.py +26 -0
  1273. vllm/ray/ray_env.py +79 -0
  1274. vllm/reasoning/__init__.py +92 -0
  1275. vllm/reasoning/abs_reasoning_parsers.py +290 -0
  1276. vllm/reasoning/basic_parsers.py +162 -0
  1277. vllm/reasoning/deepseek_r1_reasoning_parser.py +67 -0
  1278. vllm/reasoning/deepseek_v3_reasoning_parser.py +62 -0
  1279. vllm/reasoning/ernie45_reasoning_parser.py +165 -0
  1280. vllm/reasoning/glm4_moe_reasoning_parser.py +171 -0
  1281. vllm/reasoning/gptoss_reasoning_parser.py +173 -0
  1282. vllm/reasoning/granite_reasoning_parser.py +363 -0
  1283. vllm/reasoning/hunyuan_a13b_reasoning_parser.py +237 -0
  1284. vllm/reasoning/identity_reasoning_parser.py +58 -0
  1285. vllm/reasoning/minimax_m2_reasoning_parser.py +67 -0
  1286. vllm/reasoning/mistral_reasoning_parser.py +55 -0
  1287. vllm/reasoning/olmo3_reasoning_parser.py +302 -0
  1288. vllm/reasoning/qwen3_reasoning_parser.py +67 -0
  1289. vllm/reasoning/seedoss_reasoning_parser.py +27 -0
  1290. vllm/reasoning/step3_reasoning_parser.py +107 -0
  1291. vllm/sampling_params.py +669 -0
  1292. vllm/scalar_type.py +355 -0
  1293. vllm/scripts.py +17 -0
  1294. vllm/sequence.py +98 -0
  1295. vllm/tasks.py +13 -0
  1296. vllm/third_party/__init__.py +0 -0
  1297. vllm/third_party/pynvml.py +6140 -0
  1298. vllm/tracing.py +135 -0
  1299. vllm/transformers_utils/__init__.py +26 -0
  1300. vllm/transformers_utils/chat_templates/__init__.py +5 -0
  1301. vllm/transformers_utils/chat_templates/registry.py +73 -0
  1302. vllm/transformers_utils/chat_templates/template_basic.jinja +3 -0
  1303. vllm/transformers_utils/chat_templates/template_blip2.jinja +11 -0
  1304. vllm/transformers_utils/chat_templates/template_chatml.jinja +10 -0
  1305. vllm/transformers_utils/chat_templates/template_deepseek_ocr.jinja +14 -0
  1306. vllm/transformers_utils/chat_templates/template_deepseek_vl2.jinja +23 -0
  1307. vllm/transformers_utils/chat_templates/template_fuyu.jinja +3 -0
  1308. vllm/transformers_utils/chat_templates/template_minicpmv45.jinja +93 -0
  1309. vllm/transformers_utils/config.py +1203 -0
  1310. vllm/transformers_utils/config_parser_base.py +20 -0
  1311. vllm/transformers_utils/configs/__init__.py +70 -0
  1312. vllm/transformers_utils/configs/afmoe.py +84 -0
  1313. vllm/transformers_utils/configs/arctic.py +206 -0
  1314. vllm/transformers_utils/configs/chatglm.py +75 -0
  1315. vllm/transformers_utils/configs/deepseek_vl2.py +126 -0
  1316. vllm/transformers_utils/configs/dotsocr.py +71 -0
  1317. vllm/transformers_utils/configs/eagle.py +84 -0
  1318. vllm/transformers_utils/configs/falcon.py +89 -0
  1319. vllm/transformers_utils/configs/flex_olmo.py +77 -0
  1320. vllm/transformers_utils/configs/jais.py +243 -0
  1321. vllm/transformers_utils/configs/kimi_linear.py +144 -0
  1322. vllm/transformers_utils/configs/kimi_vl.py +38 -0
  1323. vllm/transformers_utils/configs/lfm2_moe.py +159 -0
  1324. vllm/transformers_utils/configs/medusa.py +65 -0
  1325. vllm/transformers_utils/configs/midashenglm.py +103 -0
  1326. vllm/transformers_utils/configs/mistral.py +174 -0
  1327. vllm/transformers_utils/configs/mlp_speculator.py +69 -0
  1328. vllm/transformers_utils/configs/moonvit.py +33 -0
  1329. vllm/transformers_utils/configs/nemotron.py +212 -0
  1330. vllm/transformers_utils/configs/nemotron_h.py +282 -0
  1331. vllm/transformers_utils/configs/olmo3.py +79 -0
  1332. vllm/transformers_utils/configs/ovis.py +182 -0
  1333. vllm/transformers_utils/configs/qwen3_next.py +274 -0
  1334. vllm/transformers_utils/configs/radio.py +89 -0
  1335. vllm/transformers_utils/configs/speculators/__init__.py +2 -0
  1336. vllm/transformers_utils/configs/speculators/algos.py +38 -0
  1337. vllm/transformers_utils/configs/speculators/base.py +114 -0
  1338. vllm/transformers_utils/configs/step3_vl.py +174 -0
  1339. vllm/transformers_utils/configs/ultravox.py +118 -0
  1340. vllm/transformers_utils/detokenizer_utils.py +198 -0
  1341. vllm/transformers_utils/dynamic_module.py +59 -0
  1342. vllm/transformers_utils/processor.py +402 -0
  1343. vllm/transformers_utils/processors/__init__.py +15 -0
  1344. vllm/transformers_utils/processors/deepseek_ocr.py +438 -0
  1345. vllm/transformers_utils/processors/deepseek_vl2.py +406 -0
  1346. vllm/transformers_utils/processors/ovis.py +453 -0
  1347. vllm/transformers_utils/processors/ovis2_5.py +468 -0
  1348. vllm/transformers_utils/runai_utils.py +104 -0
  1349. vllm/transformers_utils/s3_utils.py +95 -0
  1350. vllm/transformers_utils/tokenizer.py +293 -0
  1351. vllm/transformers_utils/tokenizer_base.py +155 -0
  1352. vllm/transformers_utils/tokenizers/__init__.py +16 -0
  1353. vllm/transformers_utils/tokenizers/mistral.py +502 -0
  1354. vllm/transformers_utils/utils.py +130 -0
  1355. vllm/triton_utils/__init__.py +19 -0
  1356. vllm/triton_utils/importing.py +103 -0
  1357. vllm/usage/__init__.py +0 -0
  1358. vllm/usage/usage_lib.py +294 -0
  1359. vllm/utils/__init__.py +82 -0
  1360. vllm/utils/argparse_utils.py +487 -0
  1361. vllm/utils/async_utils.py +303 -0
  1362. vllm/utils/cache.py +214 -0
  1363. vllm/utils/collection_utils.py +139 -0
  1364. vllm/utils/counter.py +45 -0
  1365. vllm/utils/deep_gemm.py +391 -0
  1366. vllm/utils/flashinfer.py +490 -0
  1367. vllm/utils/func_utils.py +236 -0
  1368. vllm/utils/gc_utils.py +147 -0
  1369. vllm/utils/hashing.py +63 -0
  1370. vllm/utils/import_utils.py +411 -0
  1371. vllm/utils/jsontree.py +165 -0
  1372. vllm/utils/math_utils.py +32 -0
  1373. vllm/utils/mem_constants.py +13 -0
  1374. vllm/utils/mem_utils.py +232 -0
  1375. vllm/utils/nccl.py +64 -0
  1376. vllm/utils/network_utils.py +331 -0
  1377. vllm/utils/platform_utils.py +59 -0
  1378. vllm/utils/profiling.py +56 -0
  1379. vllm/utils/registry.py +49 -0
  1380. vllm/utils/serial_utils.py +169 -0
  1381. vllm/utils/system_utils.py +229 -0
  1382. vllm/utils/tensor_schema.py +255 -0
  1383. vllm/utils/torch_utils.py +657 -0
  1384. vllm/v1/__init__.py +0 -0
  1385. vllm/v1/attention/__init__.py +0 -0
  1386. vllm/v1/attention/backends/__init__.py +0 -0
  1387. vllm/v1/attention/backends/cpu_attn.py +496 -0
  1388. vllm/v1/attention/backends/flash_attn.py +1028 -0
  1389. vllm/v1/attention/backends/flashinfer.py +1572 -0
  1390. vllm/v1/attention/backends/flex_attention.py +926 -0
  1391. vllm/v1/attention/backends/gdn_attn.py +387 -0
  1392. vllm/v1/attention/backends/linear_attn.py +74 -0
  1393. vllm/v1/attention/backends/mamba1_attn.py +165 -0
  1394. vllm/v1/attention/backends/mamba2_attn.py +354 -0
  1395. vllm/v1/attention/backends/mamba_attn.py +115 -0
  1396. vllm/v1/attention/backends/mla/__init__.py +0 -0
  1397. vllm/v1/attention/backends/mla/common.py +2031 -0
  1398. vllm/v1/attention/backends/mla/cutlass_mla.py +275 -0
  1399. vllm/v1/attention/backends/mla/flashattn_mla.py +337 -0
  1400. vllm/v1/attention/backends/mla/flashinfer_mla.py +171 -0
  1401. vllm/v1/attention/backends/mla/flashmla.py +314 -0
  1402. vllm/v1/attention/backends/mla/flashmla_sparse.py +548 -0
  1403. vllm/v1/attention/backends/mla/indexer.py +362 -0
  1404. vllm/v1/attention/backends/mla/rocm_aiter_mla.py +294 -0
  1405. vllm/v1/attention/backends/mla/triton_mla.py +171 -0
  1406. vllm/v1/attention/backends/pallas.py +436 -0
  1407. vllm/v1/attention/backends/rocm_aiter_fa.py +816 -0
  1408. vllm/v1/attention/backends/rocm_aiter_unified_attn.py +196 -0
  1409. vllm/v1/attention/backends/rocm_attn.py +362 -0
  1410. vllm/v1/attention/backends/short_conv_attn.py +105 -0
  1411. vllm/v1/attention/backends/tree_attn.py +425 -0
  1412. vllm/v1/attention/backends/triton_attn.py +373 -0
  1413. vllm/v1/attention/backends/utils.py +1116 -0
  1414. vllm/v1/attention/backends/xformers.py +417 -0
  1415. vllm/v1/core/__init__.py +0 -0
  1416. vllm/v1/core/block_pool.py +428 -0
  1417. vllm/v1/core/encoder_cache_manager.py +343 -0
  1418. vllm/v1/core/kv_cache_coordinator.py +480 -0
  1419. vllm/v1/core/kv_cache_manager.py +420 -0
  1420. vllm/v1/core/kv_cache_utils.py +1340 -0
  1421. vllm/v1/core/sched/__init__.py +0 -0
  1422. vllm/v1/core/sched/async_scheduler.py +62 -0
  1423. vllm/v1/core/sched/interface.py +181 -0
  1424. vllm/v1/core/sched/output.py +202 -0
  1425. vllm/v1/core/sched/request_queue.py +221 -0
  1426. vllm/v1/core/sched/scheduler.py +1617 -0
  1427. vllm/v1/core/sched/utils.py +72 -0
  1428. vllm/v1/core/single_type_kv_cache_manager.py +736 -0
  1429. vllm/v1/cudagraph_dispatcher.py +148 -0
  1430. vllm/v1/engine/__init__.py +206 -0
  1431. vllm/v1/engine/async_llm.py +797 -0
  1432. vllm/v1/engine/coordinator.py +377 -0
  1433. vllm/v1/engine/core.py +1420 -0
  1434. vllm/v1/engine/core_client.py +1400 -0
  1435. vllm/v1/engine/detokenizer.py +351 -0
  1436. vllm/v1/engine/exceptions.py +18 -0
  1437. vllm/v1/engine/llm_engine.py +408 -0
  1438. vllm/v1/engine/logprobs.py +182 -0
  1439. vllm/v1/engine/output_processor.py +642 -0
  1440. vllm/v1/engine/parallel_sampling.py +145 -0
  1441. vllm/v1/engine/processor.py +621 -0
  1442. vllm/v1/engine/utils.py +1072 -0
  1443. vllm/v1/executor/__init__.py +6 -0
  1444. vllm/v1/executor/abstract.py +352 -0
  1445. vllm/v1/executor/multiproc_executor.py +877 -0
  1446. vllm/v1/executor/ray_distributed_executor.py +8 -0
  1447. vllm/v1/executor/ray_executor.py +626 -0
  1448. vllm/v1/executor/ray_utils.py +465 -0
  1449. vllm/v1/executor/uniproc_executor.py +183 -0
  1450. vllm/v1/kv_cache_interface.py +403 -0
  1451. vllm/v1/kv_offload/__init__.py +0 -0
  1452. vllm/v1/kv_offload/abstract.py +161 -0
  1453. vllm/v1/kv_offload/arc_manager.py +237 -0
  1454. vllm/v1/kv_offload/backend.py +97 -0
  1455. vllm/v1/kv_offload/backends/__init__.py +0 -0
  1456. vllm/v1/kv_offload/backends/cpu.py +62 -0
  1457. vllm/v1/kv_offload/cpu.py +93 -0
  1458. vllm/v1/kv_offload/factory.py +56 -0
  1459. vllm/v1/kv_offload/lru_manager.py +139 -0
  1460. vllm/v1/kv_offload/mediums.py +39 -0
  1461. vllm/v1/kv_offload/spec.py +62 -0
  1462. vllm/v1/kv_offload/worker/__init__.py +0 -0
  1463. vllm/v1/kv_offload/worker/cpu_gpu.py +185 -0
  1464. vllm/v1/kv_offload/worker/worker.py +144 -0
  1465. vllm/v1/metrics/__init__.py +0 -0
  1466. vllm/v1/metrics/loggers.py +1238 -0
  1467. vllm/v1/metrics/prometheus.py +82 -0
  1468. vllm/v1/metrics/ray_wrappers.py +169 -0
  1469. vllm/v1/metrics/reader.py +257 -0
  1470. vllm/v1/metrics/stats.py +420 -0
  1471. vllm/v1/outputs.py +249 -0
  1472. vllm/v1/pool/__init__.py +0 -0
  1473. vllm/v1/pool/metadata.py +82 -0
  1474. vllm/v1/request.py +259 -0
  1475. vllm/v1/sample/__init__.py +0 -0
  1476. vllm/v1/sample/logits_processor/__init__.py +352 -0
  1477. vllm/v1/sample/logits_processor/builtin.py +274 -0
  1478. vllm/v1/sample/logits_processor/interface.py +106 -0
  1479. vllm/v1/sample/logits_processor/state.py +165 -0
  1480. vllm/v1/sample/metadata.py +44 -0
  1481. vllm/v1/sample/ops/__init__.py +0 -0
  1482. vllm/v1/sample/ops/bad_words.py +52 -0
  1483. vllm/v1/sample/ops/logprobs.py +25 -0
  1484. vllm/v1/sample/ops/penalties.py +57 -0
  1485. vllm/v1/sample/ops/topk_topp_sampler.py +290 -0
  1486. vllm/v1/sample/rejection_sampler.py +793 -0
  1487. vllm/v1/sample/sampler.py +316 -0
  1488. vllm/v1/sample/tpu/__init__.py +0 -0
  1489. vllm/v1/sample/tpu/metadata.py +120 -0
  1490. vllm/v1/sample/tpu/sampler.py +215 -0
  1491. vllm/v1/serial_utils.py +532 -0
  1492. vllm/v1/spec_decode/__init__.py +0 -0
  1493. vllm/v1/spec_decode/eagle.py +1225 -0
  1494. vllm/v1/spec_decode/medusa.py +73 -0
  1495. vllm/v1/spec_decode/metadata.py +66 -0
  1496. vllm/v1/spec_decode/metrics.py +224 -0
  1497. vllm/v1/spec_decode/ngram_proposer.py +291 -0
  1498. vllm/v1/spec_decode/suffix_decoding.py +103 -0
  1499. vllm/v1/spec_decode/utils.py +16 -0
  1500. vllm/v1/structured_output/__init__.py +338 -0
  1501. vllm/v1/structured_output/backend_guidance.py +265 -0
  1502. vllm/v1/structured_output/backend_lm_format_enforcer.py +177 -0
  1503. vllm/v1/structured_output/backend_outlines.py +324 -0
  1504. vllm/v1/structured_output/backend_types.py +136 -0
  1505. vllm/v1/structured_output/backend_xgrammar.py +362 -0
  1506. vllm/v1/structured_output/request.py +94 -0
  1507. vllm/v1/structured_output/utils.py +469 -0
  1508. vllm/v1/utils.py +414 -0
  1509. vllm/v1/worker/__init__.py +0 -0
  1510. vllm/v1/worker/block_table.py +327 -0
  1511. vllm/v1/worker/cpu_model_runner.py +122 -0
  1512. vllm/v1/worker/cpu_worker.py +206 -0
  1513. vllm/v1/worker/dp_utils.py +230 -0
  1514. vllm/v1/worker/ec_connector_model_runner_mixin.py +87 -0
  1515. vllm/v1/worker/gpu_input_batch.py +975 -0
  1516. vllm/v1/worker/gpu_model_runner.py +5102 -0
  1517. vllm/v1/worker/gpu_ubatch_wrapper.py +466 -0
  1518. vllm/v1/worker/gpu_worker.py +894 -0
  1519. vllm/v1/worker/kv_connector_model_runner_mixin.py +144 -0
  1520. vllm/v1/worker/lora_model_runner_mixin.py +213 -0
  1521. vllm/v1/worker/tpu_input_batch.py +593 -0
  1522. vllm/v1/worker/tpu_model_runner.py +2173 -0
  1523. vllm/v1/worker/tpu_worker.py +355 -0
  1524. vllm/v1/worker/ubatch_utils.py +73 -0
  1525. vllm/v1/worker/ubatching.py +231 -0
  1526. vllm/v1/worker/utils.py +366 -0
  1527. vllm/v1/worker/worker_base.py +375 -0
  1528. vllm/v1/worker/xpu_model_runner.py +55 -0
  1529. vllm/v1/worker/xpu_worker.py +189 -0
  1530. vllm/version.py +39 -0
  1531. vllm/vllm_flash_attn/.gitkeep +0 -0
  1532. vllm_cpu_amxbf16-0.11.2.post2.dist-info/METADATA +345 -0
  1533. vllm_cpu_amxbf16-0.11.2.post2.dist-info/RECORD +1536 -0
  1534. vllm_cpu_amxbf16-0.11.2.post2.dist-info/WHEEL +5 -0
  1535. vllm_cpu_amxbf16-0.11.2.post2.dist-info/entry_points.txt +5 -0
  1536. vllm_cpu_amxbf16-0.11.2.post2.dist-info/top_level.txt +1 -0
@@ -0,0 +1,781 @@
1
+ # SPDX-License-Identifier: Apache-2.0
2
+ # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
+ import ast
4
+ import json
5
+ import uuid
6
+ from collections.abc import Sequence
7
+ from typing import Any
8
+
9
+ import regex as re
10
+
11
+ from vllm.entrypoints.openai.protocol import (
12
+ ChatCompletionRequest,
13
+ ChatCompletionToolsParam,
14
+ DeltaFunctionCall,
15
+ DeltaMessage,
16
+ DeltaToolCall,
17
+ ExtractedToolCallInformation,
18
+ FunctionCall,
19
+ ToolCall,
20
+ )
21
+ from vllm.entrypoints.openai.tool_parsers.abstract_tool_parser import (
22
+ ToolParser,
23
+ )
24
+ from vllm.logger import init_logger
25
+ from vllm.transformers_utils.tokenizer import AnyTokenizer
26
+
27
+ logger = init_logger(__name__)
28
+
29
+
30
+ class Qwen3CoderToolParser(ToolParser):
31
+ def __init__(self, tokenizer: AnyTokenizer):
32
+ super().__init__(tokenizer)
33
+
34
+ self.current_tool_name_sent: bool = False
35
+ self.prev_tool_call_arr: list[dict] = []
36
+ # Override base class type - we use string IDs for tool calls
37
+ self.current_tool_id: str | None = None # type: ignore
38
+ self.streamed_args_for_tool: list[str] = []
39
+
40
+ # Sentinel tokens for streaming mode
41
+ self.tool_call_start_token: str = "<tool_call>"
42
+ self.tool_call_end_token: str = "</tool_call>"
43
+ self.tool_call_prefix: str = "<function="
44
+ self.function_end_token: str = "</function>"
45
+ self.parameter_prefix: str = "<parameter="
46
+ self.parameter_end_token: str = "</parameter>"
47
+ self.is_tool_call_started: bool = False
48
+ self.failed_count: int = 0
49
+
50
+ # Enhanced streaming state - reset for each new message
51
+ self._reset_streaming_state()
52
+
53
+ # Regex patterns
54
+ self.tool_call_complete_regex = re.compile(
55
+ r"<tool_call>(.*?)</tool_call>", re.DOTALL
56
+ )
57
+ self.tool_call_regex = re.compile(
58
+ r"<tool_call>(.*?)</tool_call>|<tool_call>(.*?)$", re.DOTALL
59
+ )
60
+ self.tool_call_function_regex = re.compile(
61
+ r"<function=(.*?)</function>|<function=(.*)$", re.DOTALL
62
+ )
63
+ self.tool_call_parameter_regex = re.compile(
64
+ r"<parameter=(.*?)(?:</parameter>|(?=<parameter=)|(?=</function>)|$)",
65
+ re.DOTALL,
66
+ )
67
+
68
+ if not self.model_tokenizer:
69
+ raise ValueError(
70
+ "The model tokenizer must be passed to the ToolParser "
71
+ "constructor during construction."
72
+ )
73
+
74
+ self.tool_call_start_token_id = self.vocab.get(self.tool_call_start_token)
75
+ self.tool_call_end_token_id = self.vocab.get(self.tool_call_end_token)
76
+
77
+ if self.tool_call_start_token_id is None or self.tool_call_end_token_id is None:
78
+ raise RuntimeError(
79
+ "Qwen3 XML Tool parser could not locate tool call start/end "
80
+ "tokens in the tokenizer!"
81
+ )
82
+
83
+ logger.info(
84
+ "vLLM Successfully import tool parser %s !", self.__class__.__name__
85
+ )
86
+
87
+ def _generate_tool_call_id(self) -> str:
88
+ """Generate a unique tool call ID."""
89
+ return f"call_{uuid.uuid4().hex[:24]}"
90
+
91
+ def _reset_streaming_state(self):
92
+ """Reset all streaming state."""
93
+ self.current_tool_index = 0
94
+ self.is_tool_call_started = False
95
+ self.header_sent = False
96
+ self.current_tool_id = None
97
+ self.current_function_name = None
98
+ self.current_param_name = None
99
+ self.current_param_value = ""
100
+ self.param_count = 0
101
+ self.in_param = False
102
+ self.in_function = False
103
+ self.accumulated_text = ""
104
+ self.json_started = False
105
+ self.json_closed = False
106
+ # Store accumulated parameters for type conversion
107
+ self.accumulated_params = {}
108
+ self.streaming_request = None
109
+
110
+ def _get_arguments_config(
111
+ self, func_name: str, tools: list[ChatCompletionToolsParam] | None
112
+ ) -> dict:
113
+ """Extract argument configuration for a function."""
114
+ if tools is None:
115
+ return {}
116
+ for config in tools:
117
+ if not hasattr(config, "type") or not (
118
+ hasattr(config, "function") and hasattr(config.function, "name")
119
+ ):
120
+ continue
121
+ if config.type == "function" and config.function.name == func_name:
122
+ if not hasattr(config.function, "parameters"):
123
+ return {}
124
+ params = config.function.parameters
125
+ if isinstance(params, dict) and "properties" in params:
126
+ return params["properties"]
127
+ elif isinstance(params, dict):
128
+ return params
129
+ else:
130
+ return {}
131
+ logger.warning("Tool '%s' is not defined in the tools list.", func_name)
132
+ return {}
133
+
134
+ def _convert_param_value(
135
+ self, param_value: str, param_name: str, param_config: dict, func_name: str
136
+ ) -> Any:
137
+ """Convert parameter value based on its type in the schema."""
138
+ # Handle null value for any type
139
+ if param_value.lower() == "null":
140
+ return None
141
+
142
+ if param_name not in param_config:
143
+ if param_config != {}:
144
+ logger.warning(
145
+ "Parsed parameter '%s' is not defined in the tool "
146
+ "parameters for tool '%s', directly returning the "
147
+ "string value.",
148
+ param_name,
149
+ func_name,
150
+ )
151
+ return param_value
152
+
153
+ if (
154
+ isinstance(param_config[param_name], dict)
155
+ and "type" in param_config[param_name]
156
+ ):
157
+ param_type = str(param_config[param_name]["type"]).strip().lower()
158
+ else:
159
+ param_type = "string"
160
+ if param_type in ["string", "str", "text", "varchar", "char", "enum"]:
161
+ return param_value
162
+ elif (
163
+ param_type.startswith("int")
164
+ or param_type.startswith("uint")
165
+ or param_type.startswith("long")
166
+ or param_type.startswith("short")
167
+ or param_type.startswith("unsigned")
168
+ ):
169
+ try:
170
+ return int(param_value)
171
+ except (ValueError, TypeError):
172
+ logger.warning(
173
+ "Parsed value '%s' of parameter '%s' is not an "
174
+ "integer in tool '%s', degenerating to string.",
175
+ param_value,
176
+ param_name,
177
+ func_name,
178
+ )
179
+ return param_value
180
+ elif param_type.startswith("num") or param_type.startswith("float"):
181
+ try:
182
+ float_param_value = float(param_value)
183
+ return (
184
+ float_param_value
185
+ if float_param_value - int(float_param_value) != 0
186
+ else int(float_param_value)
187
+ )
188
+ except (ValueError, TypeError):
189
+ logger.warning(
190
+ "Parsed value '%s' of parameter '%s' is not a float "
191
+ "in tool '%s', degenerating to string.",
192
+ param_value,
193
+ param_name,
194
+ func_name,
195
+ )
196
+ return param_value
197
+ elif param_type in ["boolean", "bool", "binary"]:
198
+ param_value = param_value.lower()
199
+ if param_value not in ["true", "false"]:
200
+ logger.warning(
201
+ "Parsed value '%s' of parameter '%s' is not a boolean "
202
+ "(`true` or `false`) in tool '%s', degenerating to "
203
+ "false.",
204
+ param_value,
205
+ param_name,
206
+ func_name,
207
+ )
208
+ return param_value == "true"
209
+ else:
210
+ if (
211
+ param_type in ["object", "array", "arr"]
212
+ or param_type.startswith("dict")
213
+ or param_type.startswith("list")
214
+ ):
215
+ try:
216
+ param_value = json.loads(param_value)
217
+ return param_value
218
+ except (json.JSONDecodeError, TypeError, ValueError):
219
+ logger.warning(
220
+ "Parsed value '%s' of parameter '%s' cannot be "
221
+ "parsed with json.loads in tool '%s', will try "
222
+ "other methods to parse it.",
223
+ param_value,
224
+ param_name,
225
+ func_name,
226
+ )
227
+ try:
228
+ param_value = ast.literal_eval(param_value) # safer
229
+ except (ValueError, SyntaxError, TypeError):
230
+ logger.warning(
231
+ "Parsed value '%s' of parameter '%s' cannot be "
232
+ "converted via Python `ast.literal_eval()` in tool "
233
+ "'%s', degenerating to string.",
234
+ param_value,
235
+ param_name,
236
+ func_name,
237
+ )
238
+ return param_value
239
+
240
+ def _parse_xml_function_call(
241
+ self, function_call_str: str, tools: list[ChatCompletionToolsParam] | None
242
+ ) -> ToolCall | None:
243
+ # Extract function name
244
+ end_index = function_call_str.index(">")
245
+ function_name = function_call_str[:end_index]
246
+ param_config = self._get_arguments_config(function_name, tools)
247
+ parameters = function_call_str[end_index + 1 :]
248
+ param_dict = {}
249
+ for match_text in self.tool_call_parameter_regex.findall(parameters):
250
+ idx = match_text.index(">")
251
+ param_name = match_text[:idx]
252
+ param_value = str(match_text[idx + 1 :])
253
+ # Remove prefix and trailing \n
254
+ if param_value.startswith("\n"):
255
+ param_value = param_value[1:]
256
+ if param_value.endswith("\n"):
257
+ param_value = param_value[:-1]
258
+
259
+ param_dict[param_name] = self._convert_param_value(
260
+ param_value, param_name, param_config, function_name
261
+ )
262
+ return ToolCall(
263
+ type="function",
264
+ function=FunctionCall(
265
+ name=function_name, arguments=json.dumps(param_dict, ensure_ascii=False)
266
+ ),
267
+ )
268
+
269
+ def _get_function_calls(self, model_output: str) -> list[str]:
270
+ # Find all tool calls
271
+ matched_ranges = self.tool_call_regex.findall(model_output)
272
+ raw_tool_calls = [
273
+ match[0] if match[0] else match[1] for match in matched_ranges
274
+ ]
275
+
276
+ # Back-off strategy if no tool_call tags found
277
+ if len(raw_tool_calls) == 0:
278
+ raw_tool_calls = [model_output]
279
+
280
+ raw_function_calls = []
281
+ for tool_call in raw_tool_calls:
282
+ raw_function_calls.extend(self.tool_call_function_regex.findall(tool_call))
283
+
284
+ function_calls = [
285
+ match[0] if match[0] else match[1] for match in raw_function_calls
286
+ ]
287
+ return function_calls
288
+
289
+ def extract_tool_calls(
290
+ self,
291
+ model_output: str,
292
+ request: ChatCompletionRequest,
293
+ ) -> ExtractedToolCallInformation:
294
+ # Quick check to avoid unnecessary processing
295
+ if self.tool_call_prefix not in model_output:
296
+ return ExtractedToolCallInformation(
297
+ tools_called=False, tool_calls=[], content=model_output
298
+ )
299
+
300
+ try:
301
+ function_calls = self._get_function_calls(model_output)
302
+ if len(function_calls) == 0:
303
+ return ExtractedToolCallInformation(
304
+ tools_called=False, tool_calls=[], content=model_output
305
+ )
306
+
307
+ tool_calls = [
308
+ self._parse_xml_function_call(function_call_str, request.tools)
309
+ for function_call_str in function_calls
310
+ ]
311
+
312
+ # Populate prev_tool_call_arr for serving layer to set finish_reason
313
+ self.prev_tool_call_arr.clear() # Clear previous calls
314
+ for tool_call in tool_calls:
315
+ if tool_call:
316
+ self.prev_tool_call_arr.append(
317
+ {
318
+ "name": tool_call.function.name,
319
+ "arguments": tool_call.function.arguments,
320
+ }
321
+ )
322
+
323
+ # Extract content before tool calls
324
+ content_index = model_output.find(self.tool_call_start_token)
325
+ idx = model_output.find(self.tool_call_prefix)
326
+ content_index = content_index if content_index >= 0 else idx
327
+ content = model_output[:content_index] # .rstrip()
328
+
329
+ return ExtractedToolCallInformation(
330
+ tools_called=(len(tool_calls) > 0),
331
+ tool_calls=tool_calls,
332
+ content=content if content else None,
333
+ )
334
+
335
+ except Exception:
336
+ logger.exception("Error in extracting tool call from response.")
337
+ return ExtractedToolCallInformation(
338
+ tools_called=False, tool_calls=[], content=model_output
339
+ )
340
+
341
+ def extract_tool_calls_streaming(
342
+ self,
343
+ previous_text: str,
344
+ current_text: str,
345
+ delta_text: str,
346
+ previous_token_ids: Sequence[int],
347
+ current_token_ids: Sequence[int],
348
+ delta_token_ids: Sequence[int],
349
+ request: ChatCompletionRequest,
350
+ ) -> DeltaMessage | None:
351
+ # Store request for type conversion
352
+ if not previous_text:
353
+ self._reset_streaming_state()
354
+ self.streaming_request = request
355
+
356
+ # If no delta text, return None unless it's an EOS token after tools
357
+ if not delta_text:
358
+ # Check if this is an EOS token after all tool calls are complete
359
+ # Check for tool calls in text even if is_tool_call_started
360
+ # is False (might have been reset after processing all tools)
361
+ if delta_token_ids and self.tool_call_end_token_id not in delta_token_ids:
362
+ # Count complete tool calls
363
+ complete_calls = len(
364
+ self.tool_call_complete_regex.findall(current_text)
365
+ )
366
+
367
+ # If we have completed tool calls and populated
368
+ # prev_tool_call_arr
369
+ if complete_calls > 0 and len(self.prev_tool_call_arr) > 0:
370
+ # Check if all tool calls are closed
371
+ open_calls = current_text.count(
372
+ self.tool_call_start_token
373
+ ) - current_text.count(self.tool_call_end_token)
374
+ if open_calls == 0:
375
+ # Return empty delta for finish_reason processing
376
+ return DeltaMessage(content="")
377
+ elif not self.is_tool_call_started and current_text:
378
+ # This is a regular content response that's now complete
379
+ return DeltaMessage(content="")
380
+ return None
381
+
382
+ # Update accumulated text
383
+ self.accumulated_text = current_text
384
+
385
+ # Check if we need to advance to next tool
386
+ if self.json_closed and not self.in_function:
387
+ # Check if this tool call has ended
388
+ tool_ends = current_text.count(self.tool_call_end_token)
389
+ if tool_ends > self.current_tool_index:
390
+ # This tool has ended, advance to next
391
+ self.current_tool_index += 1
392
+ self.header_sent = False
393
+ self.param_count = 0
394
+ self.json_started = False
395
+ self.json_closed = False
396
+ self.accumulated_params = {}
397
+
398
+ # Check if there are more tool calls
399
+ tool_starts = current_text.count(self.tool_call_start_token)
400
+ if self.current_tool_index >= tool_starts:
401
+ # No more tool calls
402
+ self.is_tool_call_started = False
403
+ # Continue processing next tool
404
+ return None
405
+
406
+ # Handle normal content before tool calls
407
+ if not self.is_tool_call_started:
408
+ # Check if tool call is starting
409
+ if (
410
+ self.tool_call_start_token_id in delta_token_ids
411
+ or self.tool_call_start_token in delta_text
412
+ ):
413
+ self.is_tool_call_started = True
414
+ # Return any content before the tool call
415
+ if self.tool_call_start_token in delta_text:
416
+ content_before = delta_text[
417
+ : delta_text.index(self.tool_call_start_token)
418
+ ]
419
+ if content_before:
420
+ return DeltaMessage(content=content_before)
421
+ return None
422
+ else:
423
+ # Check if we're between tool calls - skip whitespace
424
+ if (
425
+ current_text.rstrip().endswith(self.tool_call_end_token)
426
+ and delta_text.strip() == ""
427
+ ):
428
+ # We just ended a tool call, skip whitespace
429
+ return None
430
+ # Normal content, no tool call
431
+ return DeltaMessage(content=delta_text)
432
+
433
+ # Check if we're between tool calls (waiting for next one)
434
+ # Count tool calls we've seen vs processed
435
+ tool_starts_count = current_text.count(self.tool_call_start_token)
436
+ if self.current_tool_index >= tool_starts_count:
437
+ # We're past all tool calls, shouldn't be here
438
+ return None
439
+
440
+ # We're in a tool call, find the current tool call portion
441
+ # Need to find the correct tool call based on current_tool_index
442
+ tool_start_positions: list[int] = []
443
+ idx = 0
444
+ while True:
445
+ idx = current_text.find(self.tool_call_start_token, idx)
446
+ if idx == -1:
447
+ break
448
+ tool_start_positions.append(idx)
449
+ idx += len(self.tool_call_start_token)
450
+
451
+ if self.current_tool_index >= len(tool_start_positions):
452
+ # No more tool calls to process yet
453
+ return None
454
+
455
+ tool_start_idx = tool_start_positions[self.current_tool_index]
456
+ # Find where this tool call ends (or current position if not ended yet)
457
+ tool_end_idx = current_text.find(self.tool_call_end_token, tool_start_idx)
458
+ if tool_end_idx == -1:
459
+ tool_text = current_text[tool_start_idx:]
460
+ else:
461
+ tool_text = current_text[
462
+ tool_start_idx : tool_end_idx + len(self.tool_call_end_token)
463
+ ]
464
+
465
+ # Looking for function header
466
+ if not self.header_sent:
467
+ if self.tool_call_prefix in tool_text:
468
+ func_start = tool_text.find(self.tool_call_prefix) + len(
469
+ self.tool_call_prefix
470
+ )
471
+ func_end = tool_text.find(">", func_start)
472
+
473
+ if func_end != -1:
474
+ # Found complete function name
475
+ self.current_function_name = tool_text[func_start:func_end]
476
+ self.current_tool_id = self._generate_tool_call_id()
477
+ self.header_sent = True
478
+ self.in_function = True
479
+
480
+ # IMPORTANT: Add to prev_tool_call_arr immediately when
481
+ # we detect a tool call. This ensures
482
+ # finish_reason="tool_calls" even if parsing isn't complete
483
+ already_added = any(
484
+ tool.get("name") == self.current_function_name
485
+ for tool in self.prev_tool_call_arr
486
+ )
487
+ if not already_added:
488
+ self.prev_tool_call_arr.append(
489
+ {
490
+ "name": self.current_function_name,
491
+ "arguments": "{}", # Placeholder, will be updated later
492
+ }
493
+ )
494
+
495
+ # Send header with function info
496
+ return DeltaMessage(
497
+ tool_calls=[
498
+ DeltaToolCall(
499
+ index=self.current_tool_index,
500
+ id=self.current_tool_id,
501
+ function=DeltaFunctionCall(
502
+ name=self.current_function_name, arguments=""
503
+ ),
504
+ type="function",
505
+ )
506
+ ]
507
+ )
508
+ return None
509
+
510
+ # We've sent header, now handle function body
511
+ if self.in_function:
512
+ # Send opening brace if not sent yet
513
+ if not self.json_started and self.parameter_prefix not in delta_text:
514
+ self.json_started = True
515
+ return DeltaMessage(
516
+ tool_calls=[
517
+ DeltaToolCall(
518
+ index=self.current_tool_index,
519
+ function=DeltaFunctionCall(arguments="{"),
520
+ )
521
+ ]
522
+ )
523
+
524
+ # Make sure json_started is set if we're processing parameters
525
+ if not self.json_started:
526
+ self.json_started = True
527
+
528
+ # Check for function end in accumulated text
529
+ if not self.json_closed and self.function_end_token in tool_text:
530
+ # Close JSON
531
+ self.json_closed = True
532
+
533
+ # Extract complete tool call to update
534
+ # prev_tool_call_arr with final arguments
535
+ # Find the function content
536
+ func_start = tool_text.find(self.tool_call_prefix) + len(
537
+ self.tool_call_prefix
538
+ )
539
+ func_content_end = tool_text.find(self.function_end_token, func_start)
540
+ if func_content_end != -1:
541
+ func_content = tool_text[func_start:func_content_end]
542
+ # Parse to get the complete arguments
543
+ try:
544
+ parsed_tool = self._parse_xml_function_call(
545
+ func_content,
546
+ self.streaming_request.tools
547
+ if self.streaming_request
548
+ else None,
549
+ )
550
+ if parsed_tool:
551
+ # Update existing entry in
552
+ # prev_tool_call_arr with complete args
553
+ for i, tool in enumerate(self.prev_tool_call_arr):
554
+ if tool.get("name") == parsed_tool.function.name:
555
+ args = parsed_tool.function.arguments
556
+ self.prev_tool_call_arr[i]["arguments"] = args
557
+ break
558
+ except Exception:
559
+ pass # Ignore parsing errors during streaming
560
+
561
+ result = DeltaMessage(
562
+ tool_calls=[
563
+ DeltaToolCall(
564
+ index=self.current_tool_index,
565
+ function=DeltaFunctionCall(arguments="}"),
566
+ )
567
+ ]
568
+ )
569
+
570
+ # Reset state for next tool
571
+ self.in_function = False
572
+ self.json_closed = True
573
+ self.accumulated_params = {}
574
+
575
+ return result
576
+
577
+ # Look for parameters
578
+ # Find all parameter starts
579
+ param_starts = []
580
+ idx = 0
581
+ while True:
582
+ idx = tool_text.find(self.parameter_prefix, idx)
583
+ if idx == -1:
584
+ break
585
+ param_starts.append(idx)
586
+ idx += len(self.parameter_prefix)
587
+
588
+ # Check if we should start a new parameter
589
+ if (
590
+ not self.in_param
591
+ and self.param_count < len(param_starts)
592
+ and len(param_starts) > self.param_count
593
+ ):
594
+ # Process the next parameter
595
+ param_idx = param_starts[self.param_count]
596
+ param_start = param_idx + len(self.parameter_prefix)
597
+ remaining = tool_text[param_start:]
598
+
599
+ if ">" in remaining:
600
+ # We have the complete parameter name
601
+ name_end = remaining.find(">")
602
+ self.current_param_name = remaining[:name_end]
603
+
604
+ # Find the parameter value
605
+ value_start = param_start + name_end + 1
606
+ value_text = tool_text[value_start:]
607
+ if value_text.startswith("\n"):
608
+ value_text = value_text[1:]
609
+
610
+ # Find where this parameter ends
611
+ param_end_idx = value_text.find(self.parameter_end_token)
612
+ if param_end_idx == -1:
613
+ # No closing tag, look for next parameter or
614
+ # function end
615
+ next_param_idx = value_text.find(self.parameter_prefix)
616
+ func_end_idx = value_text.find(self.function_end_token)
617
+
618
+ if next_param_idx != -1 and (
619
+ func_end_idx == -1 or next_param_idx < func_end_idx
620
+ ):
621
+ param_end_idx = next_param_idx
622
+ elif func_end_idx != -1:
623
+ param_end_idx = func_end_idx
624
+ else:
625
+ # Neither found, check if tool call is complete
626
+ if self.tool_call_end_token in tool_text:
627
+ # Tool call is complete, so parameter
628
+ # must be complete too. Use all
629
+ # remaining text before function end
630
+ param_end_idx = len(value_text)
631
+ else:
632
+ # Still streaming, wait for more content
633
+ return None
634
+
635
+ if param_end_idx != -1:
636
+ # Complete parameter found
637
+ param_value = value_text[:param_end_idx]
638
+ if param_value.endswith("\n"):
639
+ param_value = param_value[:-1]
640
+
641
+ # Store raw value for later processing
642
+ self.accumulated_params[self.current_param_name] = param_value
643
+
644
+ # Get parameter configuration for type conversion
645
+ param_config = self._get_arguments_config(
646
+ self.current_function_name or "",
647
+ self.streaming_request.tools
648
+ if self.streaming_request
649
+ else None,
650
+ )
651
+
652
+ # Convert param value to appropriate type
653
+ converted_value = self._convert_param_value(
654
+ param_value,
655
+ self.current_param_name,
656
+ param_config,
657
+ self.current_function_name or "",
658
+ )
659
+
660
+ # Build JSON fragment based on the converted type
661
+ # Use json.dumps to properly serialize the value
662
+ serialized_value = json.dumps(
663
+ converted_value, ensure_ascii=False
664
+ )
665
+
666
+ if self.param_count == 0:
667
+ json_fragment = (
668
+ f'"{self.current_param_name}": {serialized_value}'
669
+ )
670
+ else:
671
+ json_fragment = (
672
+ f', "{self.current_param_name}": {serialized_value}'
673
+ )
674
+
675
+ self.param_count += 1
676
+
677
+ return DeltaMessage(
678
+ tool_calls=[
679
+ DeltaToolCall(
680
+ index=self.current_tool_index,
681
+ function=DeltaFunctionCall(arguments=json_fragment),
682
+ )
683
+ ]
684
+ )
685
+
686
+ # Continue parameter value - Not used in the current implementation
687
+ # since we process complete parameters above
688
+ if self.in_param:
689
+ if self.parameter_end_token in delta_text:
690
+ # End of parameter
691
+ end_idx = delta_text.find(self.parameter_end_token)
692
+ value_chunk = delta_text[:end_idx]
693
+
694
+ # Skip past > if at start
695
+ if not self.current_param_value and ">" in value_chunk:
696
+ gt_idx = value_chunk.find(">")
697
+ value_chunk = value_chunk[gt_idx + 1 :]
698
+
699
+ if not self.current_param_value and value_chunk.startswith("\n"):
700
+ value_chunk = value_chunk[1:]
701
+
702
+ # Store complete value
703
+ full_value = self.current_param_value + value_chunk
704
+ self.accumulated_params[self.current_param_name] = full_value
705
+
706
+ # Get parameter configuration for type conversion
707
+ param_config = self._get_arguments_config(
708
+ self.current_function_name or "",
709
+ self.streaming_request.tools
710
+ if self.streaming_request
711
+ else None,
712
+ )
713
+
714
+ # Convert the parameter value to the appropriate type
715
+ converted_value = self._convert_param_value(
716
+ full_value,
717
+ self.current_param_name or "",
718
+ param_config,
719
+ self.current_function_name or "",
720
+ )
721
+
722
+ # Serialize the converted value
723
+ serialized_value = json.dumps(converted_value, ensure_ascii=False)
724
+
725
+ # Since we've been streaming the quoted version,
726
+ # we need to close it properly
727
+ # This is complex - for now just complete the value
728
+ self.in_param = False
729
+ self.current_param_value = ""
730
+
731
+ # Just close the current parameter string
732
+ return DeltaMessage(
733
+ tool_calls=[
734
+ DeltaToolCall(
735
+ index=self.current_tool_index,
736
+ function=DeltaFunctionCall(
737
+ arguments='"'
738
+ ), # Close the string quote
739
+ )
740
+ ]
741
+ )
742
+ else:
743
+ # Continue accumulating value
744
+ value_chunk = delta_text
745
+
746
+ # Handle first chunk after param name
747
+ if not self.current_param_value and ">" in value_chunk:
748
+ gt_idx = value_chunk.find(">")
749
+ value_chunk = value_chunk[gt_idx + 1 :]
750
+
751
+ if not self.current_param_value and value_chunk.startswith("\n"):
752
+ value_chunk = value_chunk[1:]
753
+
754
+ if value_chunk:
755
+ # Stream the escaped delta
756
+ prev_escaped = (
757
+ json.dumps(self.current_param_value, ensure_ascii=False)[
758
+ 1:-1
759
+ ]
760
+ if self.current_param_value
761
+ else ""
762
+ )
763
+ self.current_param_value += value_chunk
764
+ full_escaped = json.dumps(
765
+ self.current_param_value, ensure_ascii=False
766
+ )[1:-1]
767
+ delta_escaped = full_escaped[len(prev_escaped) :]
768
+
769
+ if delta_escaped:
770
+ return DeltaMessage(
771
+ tool_calls=[
772
+ DeltaToolCall(
773
+ index=self.current_tool_index,
774
+ function=DeltaFunctionCall(
775
+ arguments=delta_escaped
776
+ ),
777
+ )
778
+ ]
779
+ )
780
+
781
+ return None