vllm-cpu 0.11.0.post2__cp312-cp312-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 (1398) hide show
  1. vllm/_C.abi3.so +0 -0
  2. vllm/__init__.py +220 -0
  3. vllm/_bc_linter.py +59 -0
  4. vllm/_custom_ops.py +2044 -0
  5. vllm/_ipex_ops.py +393 -0
  6. vllm/_version.py +34 -0
  7. vllm/assets/__init__.py +0 -0
  8. vllm/assets/audio.py +45 -0
  9. vllm/assets/base.py +41 -0
  10. vllm/assets/image.py +50 -0
  11. vllm/assets/video.py +145 -0
  12. vllm/attention/__init__.py +15 -0
  13. vllm/attention/backends/__init__.py +0 -0
  14. vllm/attention/backends/abstract.py +204 -0
  15. vllm/attention/backends/utils.py +33 -0
  16. vllm/attention/layer.py +645 -0
  17. vllm/attention/layers/__init__.py +0 -0
  18. vllm/attention/layers/chunked_local_attention.py +93 -0
  19. vllm/attention/layers/cross_attention.py +162 -0
  20. vllm/attention/layers/encoder_only_attention.py +86 -0
  21. vllm/attention/ops/__init__.py +0 -0
  22. vllm/attention/ops/chunked_prefill_paged_decode.py +405 -0
  23. vllm/attention/ops/common.py +345 -0
  24. vllm/attention/ops/flashmla.py +192 -0
  25. vllm/attention/ops/merge_attn_states.py +43 -0
  26. vllm/attention/ops/paged_attn.py +262 -0
  27. vllm/attention/ops/pallas_kv_cache_update.py +124 -0
  28. vllm/attention/ops/prefix_prefill.py +928 -0
  29. vllm/attention/ops/rocm_aiter_mla.py +104 -0
  30. vllm/attention/ops/rocm_aiter_paged_attn.py +102 -0
  31. vllm/attention/ops/triton_decode_attention.py +691 -0
  32. vllm/attention/ops/triton_flash_attention.py +984 -0
  33. vllm/attention/ops/triton_merge_attn_states.py +97 -0
  34. vllm/attention/ops/triton_reshape_and_cache_flash.py +175 -0
  35. vllm/attention/ops/triton_unified_attention.py +894 -0
  36. vllm/attention/selector.py +245 -0
  37. vllm/attention/utils/__init__.py +0 -0
  38. vllm/attention/utils/fa_utils.py +85 -0
  39. vllm/attention/utils/kv_sharing_utils.py +33 -0
  40. vllm/beam_search.py +87 -0
  41. vllm/benchmarks/__init__.py +0 -0
  42. vllm/benchmarks/datasets.py +2723 -0
  43. vllm/benchmarks/latency.py +170 -0
  44. vllm/benchmarks/lib/__init__.py +3 -0
  45. vllm/benchmarks/lib/endpoint_request_func.py +533 -0
  46. vllm/benchmarks/lib/ready_checker.py +73 -0
  47. vllm/benchmarks/lib/utils.py +80 -0
  48. vllm/benchmarks/serve.py +1358 -0
  49. vllm/benchmarks/throughput.py +696 -0
  50. vllm/collect_env.py +823 -0
  51. vllm/compilation/__init__.py +0 -0
  52. vllm/compilation/activation_quant_fusion.py +189 -0
  53. vllm/compilation/backends.py +650 -0
  54. vllm/compilation/base_static_graph.py +56 -0
  55. vllm/compilation/collective_fusion.py +1188 -0
  56. vllm/compilation/compiler_interface.py +573 -0
  57. vllm/compilation/counter.py +47 -0
  58. vllm/compilation/cuda_graph.py +199 -0
  59. vllm/compilation/cuda_piecewise_backend.py +117 -0
  60. vllm/compilation/decorators.py +400 -0
  61. vllm/compilation/fix_functionalization.py +205 -0
  62. vllm/compilation/fusion.py +383 -0
  63. vllm/compilation/fusion_attn.py +295 -0
  64. vllm/compilation/fx_utils.py +84 -0
  65. vllm/compilation/inductor_pass.py +136 -0
  66. vllm/compilation/monitor.py +57 -0
  67. vllm/compilation/noop_elimination.py +158 -0
  68. vllm/compilation/pass_manager.py +125 -0
  69. vllm/compilation/post_cleanup.py +20 -0
  70. vllm/compilation/sequence_parallelism.py +478 -0
  71. vllm/compilation/torch25_custom_graph_pass.py +42 -0
  72. vllm/compilation/vllm_inductor_pass.py +156 -0
  73. vllm/compilation/wrapper.py +136 -0
  74. vllm/config/__init__.py +814 -0
  75. vllm/config/cache.py +220 -0
  76. vllm/config/compilation.py +673 -0
  77. vllm/config/device.py +74 -0
  78. vllm/config/kv_events.py +50 -0
  79. vllm/config/kv_transfer.py +111 -0
  80. vllm/config/load.py +113 -0
  81. vllm/config/lora.py +132 -0
  82. vllm/config/model.py +1912 -0
  83. vllm/config/multimodal.py +129 -0
  84. vllm/config/observability.py +99 -0
  85. vllm/config/parallel.py +524 -0
  86. vllm/config/pooler.py +97 -0
  87. vllm/config/scheduler.py +287 -0
  88. vllm/config/speculative.py +568 -0
  89. vllm/config/speech_to_text.py +39 -0
  90. vllm/config/structured_outputs.py +64 -0
  91. vllm/config/utils.py +145 -0
  92. vllm/connections.py +186 -0
  93. vllm/device_allocator/__init__.py +0 -0
  94. vllm/device_allocator/cumem.py +311 -0
  95. vllm/distributed/__init__.py +6 -0
  96. vllm/distributed/communication_op.py +41 -0
  97. vllm/distributed/device_communicators/__init__.py +0 -0
  98. vllm/distributed/device_communicators/all2all.py +440 -0
  99. vllm/distributed/device_communicators/all_reduce_utils.py +317 -0
  100. vllm/distributed/device_communicators/base_device_communicator.py +295 -0
  101. vllm/distributed/device_communicators/cpu_communicator.py +201 -0
  102. vllm/distributed/device_communicators/cuda_communicator.py +323 -0
  103. vllm/distributed/device_communicators/cuda_wrapper.py +180 -0
  104. vllm/distributed/device_communicators/custom_all_reduce.py +311 -0
  105. vllm/distributed/device_communicators/mnnvl_compat.py +28 -0
  106. vllm/distributed/device_communicators/pynccl.py +340 -0
  107. vllm/distributed/device_communicators/pynccl_allocator.py +186 -0
  108. vllm/distributed/device_communicators/pynccl_wrapper.py +416 -0
  109. vllm/distributed/device_communicators/quick_all_reduce.py +278 -0
  110. vllm/distributed/device_communicators/ray_communicator.py +258 -0
  111. vllm/distributed/device_communicators/shm_broadcast.py +589 -0
  112. vllm/distributed/device_communicators/shm_object_storage.py +635 -0
  113. vllm/distributed/device_communicators/symm_mem.py +136 -0
  114. vllm/distributed/device_communicators/tpu_communicator.py +102 -0
  115. vllm/distributed/device_communicators/xpu_communicator.py +94 -0
  116. vllm/distributed/eplb/__init__.py +8 -0
  117. vllm/distributed/eplb/eplb_state.py +620 -0
  118. vllm/distributed/eplb/rebalance_algo.py +239 -0
  119. vllm/distributed/eplb/rebalance_execute.py +424 -0
  120. vllm/distributed/kv_events.py +362 -0
  121. vllm/distributed/kv_transfer/README.md +29 -0
  122. vllm/distributed/kv_transfer/__init__.py +13 -0
  123. vllm/distributed/kv_transfer/disagg_prefill_workflow.jpg +0 -0
  124. vllm/distributed/kv_transfer/kv_connector/__init__.py +0 -0
  125. vllm/distributed/kv_transfer/kv_connector/base.py +10 -0
  126. vllm/distributed/kv_transfer/kv_connector/factory.py +113 -0
  127. vllm/distributed/kv_transfer/kv_connector/utils.py +261 -0
  128. vllm/distributed/kv_transfer/kv_connector/v1/__init__.py +6 -0
  129. vllm/distributed/kv_transfer/kv_connector/v1/base.py +388 -0
  130. vllm/distributed/kv_transfer/kv_connector/v1/lmcache_connector.py +168 -0
  131. vllm/distributed/kv_transfer/kv_connector/v1/metrics.py +100 -0
  132. vllm/distributed/kv_transfer/kv_connector/v1/multi_connector.py +328 -0
  133. vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py +1473 -0
  134. vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py +485 -0
  135. vllm/distributed/kv_transfer/kv_connector/v1/p2p/__init__.py +0 -0
  136. vllm/distributed/kv_transfer/kv_connector/v1/p2p/p2p_nccl_connector.py +488 -0
  137. vllm/distributed/kv_transfer/kv_connector/v1/p2p/p2p_nccl_engine.py +550 -0
  138. vllm/distributed/kv_transfer/kv_connector/v1/p2p/tensor_memory_pool.py +267 -0
  139. vllm/distributed/kv_transfer/kv_connector/v1/shared_storage_connector.py +418 -0
  140. vllm/distributed/kv_transfer/kv_lookup_buffer/__init__.py +0 -0
  141. vllm/distributed/kv_transfer/kv_lookup_buffer/base.py +175 -0
  142. vllm/distributed/kv_transfer/kv_lookup_buffer/mooncake_store.py +161 -0
  143. vllm/distributed/kv_transfer/kv_lookup_buffer/simple_buffer.py +237 -0
  144. vllm/distributed/kv_transfer/kv_pipe/__init__.py +0 -0
  145. vllm/distributed/kv_transfer/kv_pipe/base.py +67 -0
  146. vllm/distributed/kv_transfer/kv_pipe/mooncake_pipe.py +290 -0
  147. vllm/distributed/kv_transfer/kv_pipe/pynccl_pipe.py +280 -0
  148. vllm/distributed/kv_transfer/kv_transfer_state.py +73 -0
  149. vllm/distributed/parallel_state.py +1532 -0
  150. vllm/distributed/tpu_distributed_utils.py +178 -0
  151. vllm/distributed/utils.py +536 -0
  152. vllm/engine/__init__.py +0 -0
  153. vllm/engine/arg_utils.py +1778 -0
  154. vllm/engine/async_llm_engine.py +6 -0
  155. vllm/engine/llm_engine.py +6 -0
  156. vllm/engine/metrics.py +577 -0
  157. vllm/engine/metrics_types.py +84 -0
  158. vllm/engine/protocol.py +333 -0
  159. vllm/entrypoints/__init__.py +0 -0
  160. vllm/entrypoints/api_server.py +178 -0
  161. vllm/entrypoints/chat_utils.py +1705 -0
  162. vllm/entrypoints/cli/__init__.py +12 -0
  163. vllm/entrypoints/cli/benchmark/__init__.py +0 -0
  164. vllm/entrypoints/cli/benchmark/base.py +25 -0
  165. vllm/entrypoints/cli/benchmark/latency.py +21 -0
  166. vllm/entrypoints/cli/benchmark/main.py +55 -0
  167. vllm/entrypoints/cli/benchmark/serve.py +21 -0
  168. vllm/entrypoints/cli/benchmark/throughput.py +21 -0
  169. vllm/entrypoints/cli/collect_env.py +36 -0
  170. vllm/entrypoints/cli/main.py +60 -0
  171. vllm/entrypoints/cli/openai.py +233 -0
  172. vllm/entrypoints/cli/run_batch.py +67 -0
  173. vllm/entrypoints/cli/serve.py +232 -0
  174. vllm/entrypoints/cli/types.py +29 -0
  175. vllm/entrypoints/constants.py +10 -0
  176. vllm/entrypoints/context.py +481 -0
  177. vllm/entrypoints/harmony_utils.py +436 -0
  178. vllm/entrypoints/launcher.py +164 -0
  179. vllm/entrypoints/llm.py +1629 -0
  180. vllm/entrypoints/logger.py +79 -0
  181. vllm/entrypoints/openai/__init__.py +0 -0
  182. vllm/entrypoints/openai/api_server.py +1953 -0
  183. vllm/entrypoints/openai/cli_args.py +288 -0
  184. vllm/entrypoints/openai/logits_processors.py +90 -0
  185. vllm/entrypoints/openai/protocol.py +2757 -0
  186. vllm/entrypoints/openai/run_batch.py +491 -0
  187. vllm/entrypoints/openai/serving_chat.py +1597 -0
  188. vllm/entrypoints/openai/serving_classification.py +173 -0
  189. vllm/entrypoints/openai/serving_completion.py +692 -0
  190. vllm/entrypoints/openai/serving_embedding.py +631 -0
  191. vllm/entrypoints/openai/serving_engine.py +992 -0
  192. vllm/entrypoints/openai/serving_models.py +288 -0
  193. vllm/entrypoints/openai/serving_pooling.py +276 -0
  194. vllm/entrypoints/openai/serving_responses.py +1709 -0
  195. vllm/entrypoints/openai/serving_score.py +479 -0
  196. vllm/entrypoints/openai/serving_tokenization.py +196 -0
  197. vllm/entrypoints/openai/serving_transcription.py +136 -0
  198. vllm/entrypoints/openai/speech_to_text.py +388 -0
  199. vllm/entrypoints/openai/tool_parsers/__init__.py +55 -0
  200. vllm/entrypoints/openai/tool_parsers/abstract_tool_parser.py +164 -0
  201. vllm/entrypoints/openai/tool_parsers/deepseekv31_tool_parser.py +367 -0
  202. vllm/entrypoints/openai/tool_parsers/deepseekv3_tool_parser.py +370 -0
  203. vllm/entrypoints/openai/tool_parsers/glm4_moe_tool_parser.py +185 -0
  204. vllm/entrypoints/openai/tool_parsers/granite_20b_fc_tool_parser.py +259 -0
  205. vllm/entrypoints/openai/tool_parsers/granite_tool_parser.py +237 -0
  206. vllm/entrypoints/openai/tool_parsers/hermes_tool_parser.py +455 -0
  207. vllm/entrypoints/openai/tool_parsers/hunyuan_a13b_tool_parser.py +372 -0
  208. vllm/entrypoints/openai/tool_parsers/internlm2_tool_parser.py +216 -0
  209. vllm/entrypoints/openai/tool_parsers/jamba_tool_parser.py +308 -0
  210. vllm/entrypoints/openai/tool_parsers/kimi_k2_tool_parser.py +377 -0
  211. vllm/entrypoints/openai/tool_parsers/llama4_pythonic_tool_parser.py +316 -0
  212. vllm/entrypoints/openai/tool_parsers/llama_tool_parser.py +269 -0
  213. vllm/entrypoints/openai/tool_parsers/longcat_tool_parser.py +39 -0
  214. vllm/entrypoints/openai/tool_parsers/minimax_tool_parser.py +816 -0
  215. vllm/entrypoints/openai/tool_parsers/mistral_tool_parser.py +369 -0
  216. vllm/entrypoints/openai/tool_parsers/openai_tool_parser.py +93 -0
  217. vllm/entrypoints/openai/tool_parsers/phi4mini_tool_parser.py +112 -0
  218. vllm/entrypoints/openai/tool_parsers/pythonic_tool_parser.py +308 -0
  219. vllm/entrypoints/openai/tool_parsers/qwen3coder_tool_parser.py +707 -0
  220. vllm/entrypoints/openai/tool_parsers/qwen3xml_tool_parser.py +1137 -0
  221. vllm/entrypoints/openai/tool_parsers/seed_oss_tool_parser.py +679 -0
  222. vllm/entrypoints/openai/tool_parsers/step3_tool_parser.py +296 -0
  223. vllm/entrypoints/openai/tool_parsers/utils.py +124 -0
  224. vllm/entrypoints/openai/tool_parsers/xlam_tool_parser.py +524 -0
  225. vllm/entrypoints/renderer.py +395 -0
  226. vllm/entrypoints/score_utils.py +232 -0
  227. vllm/entrypoints/ssl.py +75 -0
  228. vllm/entrypoints/tool.py +139 -0
  229. vllm/entrypoints/tool_server.py +206 -0
  230. vllm/entrypoints/utils.py +233 -0
  231. vllm/env_override.py +23 -0
  232. vllm/envs.py +1590 -0
  233. vllm/executor/__init__.py +0 -0
  234. vllm/executor/executor_base.py +381 -0
  235. vllm/executor/msgspec_utils.py +35 -0
  236. vllm/executor/ray_distributed_executor.py +699 -0
  237. vllm/executor/ray_utils.py +410 -0
  238. vllm/executor/uniproc_executor.py +176 -0
  239. vllm/forward_context.py +402 -0
  240. vllm/inputs/__init__.py +30 -0
  241. vllm/inputs/data.py +356 -0
  242. vllm/inputs/parse.py +151 -0
  243. vllm/inputs/preprocess.py +664 -0
  244. vllm/logger.py +229 -0
  245. vllm/logging_utils/__init__.py +10 -0
  246. vllm/logging_utils/dump_input.py +81 -0
  247. vllm/logging_utils/formatter.py +79 -0
  248. vllm/logging_utils/log_time.py +32 -0
  249. vllm/logits_process.py +119 -0
  250. vllm/logprobs.py +28 -0
  251. vllm/lora/__init__.py +0 -0
  252. vllm/lora/layers/__init__.py +34 -0
  253. vllm/lora/layers/base.py +69 -0
  254. vllm/lora/layers/base_linear.py +185 -0
  255. vllm/lora/layers/column_parallel_linear.py +609 -0
  256. vllm/lora/layers/logits_processor.py +247 -0
  257. vllm/lora/layers/qkv_x_parallel_linear.py +8 -0
  258. vllm/lora/layers/replicated_linear.py +60 -0
  259. vllm/lora/layers/row_parallel_linear.py +196 -0
  260. vllm/lora/layers/utils.py +65 -0
  261. vllm/lora/layers/vocal_parallel_embedding.py +174 -0
  262. vllm/lora/lora_weights.py +199 -0
  263. vllm/lora/models.py +816 -0
  264. vllm/lora/ops/__init__.py +0 -0
  265. vllm/lora/ops/ipex_ops/__init__.py +7 -0
  266. vllm/lora/ops/ipex_ops/lora_ops.py +44 -0
  267. vllm/lora/ops/torch_ops/__init__.py +16 -0
  268. vllm/lora/ops/torch_ops/lora_ops.py +119 -0
  269. vllm/lora/ops/triton_ops/__init__.py +12 -0
  270. vllm/lora/ops/triton_ops/kernel_utils.py +243 -0
  271. vllm/lora/ops/triton_ops/lora_expand_op.py +289 -0
  272. vllm/lora/ops/triton_ops/lora_kernel_metadata.py +148 -0
  273. vllm/lora/ops/triton_ops/lora_shrink_op.py +243 -0
  274. vllm/lora/ops/triton_ops/utils.py +126 -0
  275. vllm/lora/ops/xla_ops/__init__.py +7 -0
  276. vllm/lora/ops/xla_ops/lora_ops.py +144 -0
  277. vllm/lora/peft_helper.py +127 -0
  278. vllm/lora/punica_wrapper/__init__.py +10 -0
  279. vllm/lora/punica_wrapper/punica_base.py +458 -0
  280. vllm/lora/punica_wrapper/punica_cpu.py +349 -0
  281. vllm/lora/punica_wrapper/punica_gpu.py +272 -0
  282. vllm/lora/punica_wrapper/punica_selector.py +20 -0
  283. vllm/lora/punica_wrapper/punica_tpu.py +391 -0
  284. vllm/lora/punica_wrapper/punica_xpu.py +276 -0
  285. vllm/lora/punica_wrapper/utils.py +136 -0
  286. vllm/lora/request.py +97 -0
  287. vllm/lora/resolver.py +85 -0
  288. vllm/lora/utils.py +246 -0
  289. vllm/lora/worker_manager.py +267 -0
  290. vllm/model_executor/__init__.py +12 -0
  291. vllm/model_executor/custom_op.py +194 -0
  292. vllm/model_executor/layers/__init__.py +0 -0
  293. vllm/model_executor/layers/activation.py +575 -0
  294. vllm/model_executor/layers/attention_layer_base.py +23 -0
  295. vllm/model_executor/layers/fla/__init__.py +8 -0
  296. vllm/model_executor/layers/fla/ops/__init__.py +17 -0
  297. vllm/model_executor/layers/fla/ops/chunk.py +225 -0
  298. vllm/model_executor/layers/fla/ops/chunk_delta_h.py +290 -0
  299. vllm/model_executor/layers/fla/ops/chunk_o.py +177 -0
  300. vllm/model_executor/layers/fla/ops/chunk_scaled_dot_kkt.py +140 -0
  301. vllm/model_executor/layers/fla/ops/cumsum.py +226 -0
  302. vllm/model_executor/layers/fla/ops/fused_recurrent.py +366 -0
  303. vllm/model_executor/layers/fla/ops/index.py +39 -0
  304. vllm/model_executor/layers/fla/ops/l2norm.py +143 -0
  305. vllm/model_executor/layers/fla/ops/layernorm_guard.py +337 -0
  306. vllm/model_executor/layers/fla/ops/op.py +39 -0
  307. vllm/model_executor/layers/fla/ops/solve_tril.py +365 -0
  308. vllm/model_executor/layers/fla/ops/utils.py +180 -0
  309. vllm/model_executor/layers/fla/ops/wy_fast.py +114 -0
  310. vllm/model_executor/layers/fused_moe/__init__.py +89 -0
  311. vllm/model_executor/layers/fused_moe/batched_deep_gemm_moe.py +322 -0
  312. vllm/model_executor/layers/fused_moe/batched_triton_or_deep_gemm_moe.py +141 -0
  313. vllm/model_executor/layers/fused_moe/config.py +804 -0
  314. vllm/model_executor/layers/fused_moe/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
  315. vllm/model_executor/layers/fused_moe/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  316. vllm/model_executor/layers/fused_moe/configs/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
  317. vllm/model_executor/layers/fused_moe/configs/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
  318. vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
  319. vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +218 -0
  320. vllm/model_executor/layers/fused_moe/configs/E=1,N=3072,device_name=NVIDIA_H100_80GB_HBM3.json +218 -0
  321. vllm/model_executor/layers/fused_moe/configs/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
  322. vllm/model_executor/layers/fused_moe/configs/E=1,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
  323. vllm/model_executor/layers/fused_moe/configs/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
  324. vllm/model_executor/layers/fused_moe/configs/E=1,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
  325. vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  326. vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=AMD_Instinct_MI300X.json +200 -0
  327. vllm/model_executor/layers/fused_moe/configs/E=128,N=1024,device_name=NVIDIA_H100,dtype=fp8_w8a8.json +123 -0
  328. vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  329. vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  330. vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H20-3e.json +146 -0
  331. vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H20.json +146 -0
  332. vllm/model_executor/layers/fused_moe/configs/E=128,N=192,device_name=NVIDIA_H200.json +146 -0
  333. vllm/model_executor/layers/fused_moe/configs/E=128,N=352,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +122 -0
  334. 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
  335. 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
  336. 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
  337. 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
  338. 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
  339. vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H20-3e.json +146 -0
  340. vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H20.json +146 -0
  341. 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
  342. vllm/model_executor/layers/fused_moe/configs/E=128,N=384,device_name=NVIDIA_H200.json +146 -0
  343. vllm/model_executor/layers/fused_moe/configs/E=128,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  344. vllm/model_executor/layers/fused_moe/configs/E=128,N=704,device_name=NVIDIA_B200,dtype=fp8_w8a8.json +146 -0
  345. vllm/model_executor/layers/fused_moe/configs/E=128,N=704,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +114 -0
  346. 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
  347. 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
  348. 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
  349. 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
  350. 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
  351. vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_H20.json +146 -0
  352. 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
  353. vllm/model_executor/layers/fused_moe/configs/E=128,N=768,device_name=NVIDIA_H200.json +146 -0
  354. vllm/model_executor/layers/fused_moe/configs/E=128,N=96,device_name=NVIDIA_H20.json +146 -0
  355. vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=AMD_Instinct_MI300X.json +200 -0
  356. vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_B200,dtype=fp8_w8a8.json +147 -0
  357. vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_B200.json +146 -0
  358. vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_H100.json +146 -0
  359. vllm/model_executor/layers/fused_moe/configs/E=16,N=1344,device_name=NVIDIA_A100-SXM4-40GB.json +146 -0
  360. vllm/model_executor/layers/fused_moe/configs/E=16,N=1344,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  361. vllm/model_executor/layers/fused_moe/configs/E=16,N=1344,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  362. vllm/model_executor/layers/fused_moe/configs/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
  363. vllm/model_executor/layers/fused_moe/configs/E=16,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  364. vllm/model_executor/layers/fused_moe/configs/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +218 -0
  365. vllm/model_executor/layers/fused_moe/configs/E=16,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
  366. vllm/model_executor/layers/fused_moe/configs/E=16,N=2688,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  367. vllm/model_executor/layers/fused_moe/configs/E=16,N=2688,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  368. vllm/model_executor/layers/fused_moe/configs/E=16,N=3072,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
  369. vllm/model_executor/layers/fused_moe/configs/E=16,N=3072,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  370. vllm/model_executor/layers/fused_moe/configs/E=16,N=3200,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +130 -0
  371. vllm/model_executor/layers/fused_moe/configs/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
  372. vllm/model_executor/layers/fused_moe/configs/E=16,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json +218 -0
  373. vllm/model_executor/layers/fused_moe/configs/E=16,N=6400,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +130 -0
  374. vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json +146 -0
  375. vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  376. vllm/model_executor/layers/fused_moe/configs/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json +146 -0
  377. vllm/model_executor/layers/fused_moe/configs/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +130 -0
  378. vllm/model_executor/layers/fused_moe/configs/E=160,N=192,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
  379. vllm/model_executor/layers/fused_moe/configs/E=160,N=192,device_name=NVIDIA_H20-3e.json +146 -0
  380. vllm/model_executor/layers/fused_moe/configs/E=160,N=320,device_name=NVIDIA_H20-3e.json +146 -0
  381. vllm/model_executor/layers/fused_moe/configs/E=160,N=384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  382. 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
  383. 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
  384. 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
  385. 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
  386. 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
  387. 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
  388. 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
  389. vllm/model_executor/layers/fused_moe/configs/E=256,N=1024,device_name=AMD_Instinct_MI325X,block_shape=[128,128].json +200 -0
  390. 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
  391. 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
  392. vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8.json +146 -0
  393. 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
  394. vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8.json +146 -0
  395. 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
  396. 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
  397. 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
  398. 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
  399. 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
  400. 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
  401. 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
  402. 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
  403. 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
  404. 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
  405. 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
  406. 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
  407. 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
  408. vllm/model_executor/layers/fused_moe/configs/E=256,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  409. vllm/model_executor/layers/fused_moe/configs/E=256,N=64,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
  410. 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
  411. 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
  412. 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
  413. 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
  414. 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
  415. 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
  416. 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
  417. 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
  418. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_B200.json +146 -0
  419. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_GB200,dtype=fp8_w8a8.json +147 -0
  420. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  421. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_H20-3e.json +146 -0
  422. vllm/model_executor/layers/fused_moe/configs/E=512,N=128,device_name=NVIDIA_H200.json +146 -0
  423. vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_B200.json +146 -0
  424. vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_GB200,dtype=fp8_w8a8.json +146 -0
  425. vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  426. vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_H20-3e.json +146 -0
  427. vllm/model_executor/layers/fused_moe/configs/E=512,N=256,device_name=NVIDIA_H200.json +146 -0
  428. vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_B200.json +146 -0
  429. vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_GB200,dtype=fp8_w8a8.json +146 -0
  430. vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  431. vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_H20-3e.json +146 -0
  432. vllm/model_executor/layers/fused_moe/configs/E=512,N=512,device_name=NVIDIA_H200.json +146 -0
  433. vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_B200.json +146 -0
  434. vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_H20-3e.json +146 -0
  435. vllm/model_executor/layers/fused_moe/configs/E=512,N=64,device_name=NVIDIA_H200.json +146 -0
  436. vllm/model_executor/layers/fused_moe/configs/E=60,N=1408,device_name=AMD_Instinct_MI300X.json +200 -0
  437. vllm/model_executor/layers/fused_moe/configs/E=60,N=176,device_name=AMD_Instinct_MI300X.json +200 -0
  438. vllm/model_executor/layers/fused_moe/configs/E=60,N=352,device_name=AMD_Instinct_MI300X.json +200 -0
  439. vllm/model_executor/layers/fused_moe/configs/E=60,N=704,device_name=AMD_Instinct_MI300X.json +200 -0
  440. vllm/model_executor/layers/fused_moe/configs/E=62,N=128,device_name=AMD_Instinct_MI300X.json +200 -0
  441. vllm/model_executor/layers/fused_moe/configs/E=62,N=256,device_name=AMD_Instinct_MI300X.json +200 -0
  442. vllm/model_executor/layers/fused_moe/configs/E=62,N=256,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  443. vllm/model_executor/layers/fused_moe/configs/E=62,N=512,device_name=AMD_Instinct_MI300X.json +200 -0
  444. vllm/model_executor/layers/fused_moe/configs/E=62,N=512,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  445. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  446. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
  447. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  448. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  449. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  450. vllm/model_executor/layers/fused_moe/configs/E=64,N=1280,device_name=NVIDIA_H200.json +146 -0
  451. vllm/model_executor/layers/fused_moe/configs/E=64,N=1536,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
  452. vllm/model_executor/layers/fused_moe/configs/E=64,N=2560,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  453. vllm/model_executor/layers/fused_moe/configs/E=64,N=2560,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  454. vllm/model_executor/layers/fused_moe/configs/E=64,N=2560,device_name=NVIDIA_H200.json +146 -0
  455. vllm/model_executor/layers/fused_moe/configs/E=64,N=3072,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
  456. vllm/model_executor/layers/fused_moe/configs/E=64,N=3072,device_name=NVIDIA_H20.json +146 -0
  457. vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  458. vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  459. vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  460. vllm/model_executor/layers/fused_moe/configs/E=64,N=320,device_name=NVIDIA_H200.json +146 -0
  461. vllm/model_executor/layers/fused_moe/configs/E=64,N=384,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
  462. vllm/model_executor/layers/fused_moe/configs/E=64,N=384,device_name=NVIDIA_H20.json +146 -0
  463. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  464. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_A800-SXM4-80GB.json +146 -0
  465. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json +146 -0
  466. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  467. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  468. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  469. vllm/model_executor/layers/fused_moe/configs/E=64,N=640,device_name=NVIDIA_H200.json +146 -0
  470. vllm/model_executor/layers/fused_moe/configs/E=64,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8.json +146 -0
  471. vllm/model_executor/layers/fused_moe/configs/E=64,N=768,device_name=NVIDIA_H20.json +146 -0
  472. vllm/model_executor/layers/fused_moe/configs/E=64,N=896,device_name=NVIDIA_H20.json +146 -0
  473. vllm/model_executor/layers/fused_moe/configs/E=72,N=192,device_name=AMD_Instinct_MI300X.json +200 -0
  474. vllm/model_executor/layers/fused_moe/configs/E=72,N=384,device_name=AMD_Instinct_MI300X.json +200 -0
  475. vllm/model_executor/layers/fused_moe/configs/E=72,N=384,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  476. vllm/model_executor/layers/fused_moe/configs/E=72,N=768,device_name=AMD_Instinct_MI300X.json +200 -0
  477. vllm/model_executor/layers/fused_moe/configs/E=72,N=768,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  478. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  479. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json +200 -0
  480. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  481. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI325X.json +200 -0
  482. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +138 -0
  483. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  484. vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=NVIDIA_H200.json +146 -0
  485. vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  486. vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI300X.json +200 -0
  487. vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  488. vllm/model_executor/layers/fused_moe/configs/E=8,N=16384,device_name=AMD_Instinct_MI325X.json +200 -0
  489. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  490. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json +200 -0
  491. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  492. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI325X.json +200 -0
  493. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_A100-SXM4-40GB.json +146 -0
  494. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  495. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  496. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  497. vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=NVIDIA_H200.json +146 -0
  498. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  499. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI300X.json +200 -0
  500. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  501. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=AMD_Instinct_MI325X.json +200 -0
  502. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  503. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  504. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  505. 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
  506. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  507. vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=NVIDIA_H200.json +146 -0
  508. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  509. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json +200 -0
  510. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  511. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI325X.json +200 -0
  512. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json +146 -0
  513. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  514. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_GeForce_RTX_4090,dtype=fp8_w8a8.json +146 -0
  515. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  516. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  517. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  518. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_H200.json +146 -0
  519. vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_L40S.json +173 -0
  520. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  521. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI300X.json +200 -0
  522. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  523. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=AMD_Instinct_MI325X.json +200 -0
  524. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  525. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  526. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  527. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  528. vllm/model_executor/layers/fused_moe/configs/E=8,N=4096,device_name=NVIDIA_H200.json +146 -0
  529. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  530. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json +200 -0
  531. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  532. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI325X.json +200 -0
  533. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json +146 -0
  534. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  535. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  536. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  537. vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=NVIDIA_H200.json +146 -0
  538. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json +164 -0
  539. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI300X.json +200 -0
  540. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +164 -0
  541. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=AMD_Instinct_MI325X.json +200 -0
  542. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json +146 -0
  543. vllm/model_executor/layers/fused_moe/configs/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json +146 -0
  544. vllm/model_executor/layers/fused_moe/configs/README +12 -0
  545. vllm/model_executor/layers/fused_moe/cpu_fused_moe.py +300 -0
  546. vllm/model_executor/layers/fused_moe/cutlass_moe.py +957 -0
  547. vllm/model_executor/layers/fused_moe/deep_gemm_moe.py +362 -0
  548. vllm/model_executor/layers/fused_moe/deep_gemm_utils.py +413 -0
  549. vllm/model_executor/layers/fused_moe/deepep_ht_prepare_finalize.py +361 -0
  550. vllm/model_executor/layers/fused_moe/deepep_ll_prepare_finalize.py +274 -0
  551. vllm/model_executor/layers/fused_moe/flashinfer_cutlass_moe.py +268 -0
  552. vllm/model_executor/layers/fused_moe/flashinfer_cutlass_prepare_finalize.py +300 -0
  553. vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py +184 -0
  554. vllm/model_executor/layers/fused_moe/fused_batched_moe.py +993 -0
  555. vllm/model_executor/layers/fused_moe/fused_marlin_moe.py +239 -0
  556. vllm/model_executor/layers/fused_moe/fused_moe.py +1890 -0
  557. vllm/model_executor/layers/fused_moe/gpt_oss_triton_kernels_moe.py +307 -0
  558. vllm/model_executor/layers/fused_moe/layer.py +2195 -0
  559. vllm/model_executor/layers/fused_moe/modular_kernel.py +1038 -0
  560. vllm/model_executor/layers/fused_moe/moe_align_block_size.py +87 -0
  561. vllm/model_executor/layers/fused_moe/moe_pallas.py +80 -0
  562. vllm/model_executor/layers/fused_moe/moe_permute_unpermute.py +205 -0
  563. vllm/model_executor/layers/fused_moe/moe_torch_iterative.py +60 -0
  564. vllm/model_executor/layers/fused_moe/pplx_prepare_finalize.py +341 -0
  565. vllm/model_executor/layers/fused_moe/prepare_finalize.py +70 -0
  566. vllm/model_executor/layers/fused_moe/rocm_aiter_fused_moe.py +424 -0
  567. vllm/model_executor/layers/fused_moe/routing_simulator.py +291 -0
  568. vllm/model_executor/layers/fused_moe/topk_weight_and_reduce.py +146 -0
  569. vllm/model_executor/layers/fused_moe/triton_deep_gemm_moe.py +143 -0
  570. vllm/model_executor/layers/fused_moe/trtllm_moe.py +191 -0
  571. vllm/model_executor/layers/fused_moe/utils.py +274 -0
  572. vllm/model_executor/layers/layernorm.py +395 -0
  573. vllm/model_executor/layers/lightning_attn.py +661 -0
  574. vllm/model_executor/layers/linear.py +1603 -0
  575. vllm/model_executor/layers/logits_processor.py +106 -0
  576. vllm/model_executor/layers/mamba/__init__.py +0 -0
  577. vllm/model_executor/layers/mamba/abstract.py +42 -0
  578. vllm/model_executor/layers/mamba/linear_attn.py +403 -0
  579. vllm/model_executor/layers/mamba/mamba_mixer.py +466 -0
  580. vllm/model_executor/layers/mamba/mamba_mixer2.py +764 -0
  581. vllm/model_executor/layers/mamba/mamba_utils.py +186 -0
  582. vllm/model_executor/layers/mamba/ops/__init__.py +0 -0
  583. vllm/model_executor/layers/mamba/ops/causal_conv1d.py +1092 -0
  584. vllm/model_executor/layers/mamba/ops/layernorm_gated.py +168 -0
  585. vllm/model_executor/layers/mamba/ops/mamba_ssm.py +414 -0
  586. vllm/model_executor/layers/mamba/ops/ssd_bmm.py +242 -0
  587. vllm/model_executor/layers/mamba/ops/ssd_chunk_scan.py +527 -0
  588. vllm/model_executor/layers/mamba/ops/ssd_chunk_state.py +724 -0
  589. vllm/model_executor/layers/mamba/ops/ssd_combined.py +238 -0
  590. vllm/model_executor/layers/mamba/ops/ssd_state_passing.py +200 -0
  591. vllm/model_executor/layers/mamba/short_conv.py +253 -0
  592. vllm/model_executor/layers/mla.py +173 -0
  593. vllm/model_executor/layers/pooler.py +719 -0
  594. vllm/model_executor/layers/quantization/__init__.py +157 -0
  595. vllm/model_executor/layers/quantization/auto_round.py +388 -0
  596. vllm/model_executor/layers/quantization/awq.py +228 -0
  597. vllm/model_executor/layers/quantization/awq_marlin.py +554 -0
  598. vllm/model_executor/layers/quantization/awq_triton.py +320 -0
  599. vllm/model_executor/layers/quantization/base_config.py +170 -0
  600. vllm/model_executor/layers/quantization/bitblas.py +464 -0
  601. vllm/model_executor/layers/quantization/bitsandbytes.py +627 -0
  602. vllm/model_executor/layers/quantization/compressed_tensors/__init__.py +0 -0
  603. vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py +797 -0
  604. vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py +2074 -0
  605. vllm/model_executor/layers/quantization/compressed_tensors/schemes/__init__.py +27 -0
  606. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_24.py +366 -0
  607. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_scheme.py +55 -0
  608. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a16_24.py +160 -0
  609. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a16_nvfp4.py +105 -0
  610. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_nvfp4.py +185 -0
  611. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a8_fp8.py +169 -0
  612. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a8_int.py +135 -0
  613. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a16_fp8.py +121 -0
  614. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py +157 -0
  615. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py +111 -0
  616. vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16.py +201 -0
  617. vllm/model_executor/layers/quantization/compressed_tensors/transform/__init__.py +0 -0
  618. vllm/model_executor/layers/quantization/compressed_tensors/transform/linear.py +238 -0
  619. vllm/model_executor/layers/quantization/compressed_tensors/transform/module.py +153 -0
  620. vllm/model_executor/layers/quantization/compressed_tensors/transform/schemes/__init__.py +0 -0
  621. vllm/model_executor/layers/quantization/compressed_tensors/transform/schemes/linear_qutlass_nvfp4.py +46 -0
  622. vllm/model_executor/layers/quantization/compressed_tensors/transform/utils.py +13 -0
  623. vllm/model_executor/layers/quantization/compressed_tensors/triton_scaled_mm.py +206 -0
  624. vllm/model_executor/layers/quantization/compressed_tensors/utils.py +216 -0
  625. vllm/model_executor/layers/quantization/deepspeedfp.py +196 -0
  626. vllm/model_executor/layers/quantization/experts_int8.py +223 -0
  627. vllm/model_executor/layers/quantization/fbgemm_fp8.py +172 -0
  628. vllm/model_executor/layers/quantization/fp8.py +1098 -0
  629. vllm/model_executor/layers/quantization/gguf.py +599 -0
  630. vllm/model_executor/layers/quantization/gptq.py +340 -0
  631. vllm/model_executor/layers/quantization/gptq_bitblas.py +448 -0
  632. vllm/model_executor/layers/quantization/gptq_marlin.py +751 -0
  633. vllm/model_executor/layers/quantization/gptq_marlin_24.py +297 -0
  634. vllm/model_executor/layers/quantization/hqq_marlin.py +333 -0
  635. vllm/model_executor/layers/quantization/inc.py +61 -0
  636. vllm/model_executor/layers/quantization/input_quant_fp8.py +156 -0
  637. vllm/model_executor/layers/quantization/ipex_quant.py +415 -0
  638. vllm/model_executor/layers/quantization/kernels/__init__.py +0 -0
  639. vllm/model_executor/layers/quantization/kernels/mixed_precision/MPLinearKernel.py +91 -0
  640. vllm/model_executor/layers/quantization/kernels/mixed_precision/__init__.py +93 -0
  641. vllm/model_executor/layers/quantization/kernels/mixed_precision/allspark.py +116 -0
  642. vllm/model_executor/layers/quantization/kernels/mixed_precision/bitblas.py +302 -0
  643. vllm/model_executor/layers/quantization/kernels/mixed_precision/conch.py +92 -0
  644. vllm/model_executor/layers/quantization/kernels/mixed_precision/cutlass.py +117 -0
  645. vllm/model_executor/layers/quantization/kernels/mixed_precision/dynamic_4bit.py +92 -0
  646. vllm/model_executor/layers/quantization/kernels/mixed_precision/exllama.py +143 -0
  647. vllm/model_executor/layers/quantization/kernels/mixed_precision/machete.py +144 -0
  648. vllm/model_executor/layers/quantization/kernels/mixed_precision/marlin.py +139 -0
  649. vllm/model_executor/layers/quantization/kernels/scaled_mm/ScaledMMLinearKernel.py +67 -0
  650. vllm/model_executor/layers/quantization/kernels/scaled_mm/__init__.py +89 -0
  651. vllm/model_executor/layers/quantization/kernels/scaled_mm/aiter.py +161 -0
  652. vllm/model_executor/layers/quantization/kernels/scaled_mm/cpu.py +206 -0
  653. vllm/model_executor/layers/quantization/kernels/scaled_mm/cutlass.py +137 -0
  654. vllm/model_executor/layers/quantization/kernels/scaled_mm/triton.py +41 -0
  655. vllm/model_executor/layers/quantization/kernels/scaled_mm/xla.py +104 -0
  656. vllm/model_executor/layers/quantization/kv_cache.py +143 -0
  657. vllm/model_executor/layers/quantization/modelopt.py +1596 -0
  658. vllm/model_executor/layers/quantization/moe_wna16.py +484 -0
  659. vllm/model_executor/layers/quantization/mxfp4.py +988 -0
  660. vllm/model_executor/layers/quantization/petit.py +306 -0
  661. vllm/model_executor/layers/quantization/ptpc_fp8.py +129 -0
  662. vllm/model_executor/layers/quantization/quark/__init__.py +0 -0
  663. vllm/model_executor/layers/quantization/quark/quark.py +432 -0
  664. vllm/model_executor/layers/quantization/quark/quark_moe.py +561 -0
  665. vllm/model_executor/layers/quantization/quark/schemes/__init__.py +9 -0
  666. vllm/model_executor/layers/quantization/quark/schemes/quark_scheme.py +55 -0
  667. vllm/model_executor/layers/quantization/quark/schemes/quark_w4a4_mxfp4.py +239 -0
  668. vllm/model_executor/layers/quantization/quark/schemes/quark_w8a8_fp8.py +163 -0
  669. vllm/model_executor/layers/quantization/quark/schemes/quark_w8a8_int8.py +122 -0
  670. vllm/model_executor/layers/quantization/quark/utils.py +105 -0
  671. vllm/model_executor/layers/quantization/rtn.py +466 -0
  672. vllm/model_executor/layers/quantization/schema.py +86 -0
  673. vllm/model_executor/layers/quantization/torchao.py +214 -0
  674. vllm/model_executor/layers/quantization/tpu_int8.py +125 -0
  675. vllm/model_executor/layers/quantization/utils/__init__.py +6 -0
  676. vllm/model_executor/layers/quantization/utils/allspark_utils.py +52 -0
  677. vllm/model_executor/layers/quantization/utils/bitblas_utils.py +210 -0
  678. 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
  679. 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
  680. 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
  681. 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
  682. 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
  683. 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
  684. 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
  685. 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
  686. 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
  687. 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
  688. 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
  689. 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
  690. 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
  691. 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
  692. 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
  693. 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
  694. 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
  695. 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
  696. 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
  697. 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
  698. 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
  699. 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
  700. 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
  701. 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
  702. 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
  703. 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
  704. 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
  705. 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
  706. 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
  707. 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
  708. 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
  709. 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
  710. 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
  711. 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
  712. 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
  713. 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
  714. 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
  715. 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
  716. 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
  717. 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
  718. 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
  719. 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
  720. 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
  721. 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
  722. 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
  723. 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
  724. 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
  725. 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
  726. 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
  727. 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
  728. 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
  729. 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
  730. 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
  731. 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
  732. 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
  733. 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
  734. 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
  735. 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
  736. 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
  737. 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
  738. 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
  739. 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
  740. 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
  741. 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
  742. 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
  743. 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
  744. 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
  745. 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
  746. 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
  747. 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
  748. 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
  749. 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
  750. 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
  751. 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
  752. 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
  753. 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
  754. 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
  755. 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
  756. 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
  757. 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
  758. 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
  759. 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
  760. 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
  761. 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
  762. 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
  763. 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
  764. 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
  765. 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
  766. 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
  767. 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
  768. 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
  769. 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
  770. 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
  771. 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
  772. 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
  773. 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
  774. 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
  775. 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
  776. 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
  777. 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
  778. 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
  779. 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
  780. 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
  781. 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
  782. 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
  783. 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
  784. 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
  785. 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
  786. 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
  787. 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
  788. 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
  789. 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
  790. 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
  791. 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
  792. 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
  793. 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
  794. 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
  795. 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
  796. 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
  797. 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
  798. 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
  799. 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
  800. 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
  801. 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
  802. 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
  803. 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
  804. 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
  805. 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
  806. 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
  807. 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
  808. 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
  809. 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
  810. 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
  811. 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
  812. 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
  813. 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
  814. 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
  815. 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
  816. 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
  817. 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
  818. 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
  819. 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
  820. 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
  821. 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
  822. 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
  823. 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
  824. 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
  825. 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
  826. 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
  827. 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
  828. 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
  829. 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
  830. 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
  831. 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
  832. 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
  833. 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
  834. 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
  835. 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
  836. 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
  837. 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
  838. 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
  839. 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
  840. 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
  841. 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
  842. 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
  843. 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
  844. 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
  845. 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
  846. 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
  847. 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
  848. 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
  849. 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
  850. 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
  851. 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
  852. 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
  853. 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
  854. 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
  855. 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
  856. 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
  857. 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
  858. 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
  859. 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
  860. 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
  861. 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
  862. 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
  863. 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
  864. 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
  865. 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
  866. 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
  867. 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
  868. 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
  869. 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
  870. 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
  871. 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
  872. 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
  873. 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
  874. 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
  875. 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
  876. 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
  877. 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
  878. 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
  879. 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
  880. 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
  881. 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
  882. 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
  883. 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
  884. 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
  885. 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
  886. 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
  887. 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
  888. vllm/model_executor/layers/quantization/utils/configs/README.md +3 -0
  889. vllm/model_executor/layers/quantization/utils/flashinfer_fp4_moe.py +79 -0
  890. vllm/model_executor/layers/quantization/utils/flashinfer_utils.py +248 -0
  891. vllm/model_executor/layers/quantization/utils/fp8_utils.py +949 -0
  892. vllm/model_executor/layers/quantization/utils/gptq_utils.py +146 -0
  893. vllm/model_executor/layers/quantization/utils/int8_utils.py +492 -0
  894. vllm/model_executor/layers/quantization/utils/layer_utils.py +40 -0
  895. vllm/model_executor/layers/quantization/utils/machete_utils.py +50 -0
  896. vllm/model_executor/layers/quantization/utils/marlin_utils.py +479 -0
  897. vllm/model_executor/layers/quantization/utils/marlin_utils_fp4.py +396 -0
  898. vllm/model_executor/layers/quantization/utils/marlin_utils_fp8.py +345 -0
  899. vllm/model_executor/layers/quantization/utils/marlin_utils_test.py +165 -0
  900. vllm/model_executor/layers/quantization/utils/marlin_utils_test_24.py +464 -0
  901. vllm/model_executor/layers/quantization/utils/mxfp4_utils.py +141 -0
  902. vllm/model_executor/layers/quantization/utils/mxfp8_utils.py +20 -0
  903. vllm/model_executor/layers/quantization/utils/nvfp4_emulation_utils.py +137 -0
  904. vllm/model_executor/layers/quantization/utils/nvfp4_moe_support.py +59 -0
  905. vllm/model_executor/layers/quantization/utils/petit_utils.py +122 -0
  906. vllm/model_executor/layers/quantization/utils/quant_utils.py +641 -0
  907. vllm/model_executor/layers/quantization/utils/w8a8_utils.py +458 -0
  908. vllm/model_executor/layers/resampler.py +270 -0
  909. vllm/model_executor/layers/rotary_embedding/__init__.py +204 -0
  910. vllm/model_executor/layers/rotary_embedding/base.py +177 -0
  911. vllm/model_executor/layers/rotary_embedding/common.py +150 -0
  912. vllm/model_executor/layers/rotary_embedding/deepseek_scaling_rope.py +138 -0
  913. vllm/model_executor/layers/rotary_embedding/dual_chunk_rope.py +197 -0
  914. vllm/model_executor/layers/rotary_embedding/dynamic_ntk_alpha_rope.py +41 -0
  915. vllm/model_executor/layers/rotary_embedding/dynamic_ntk_scaling_rope.py +67 -0
  916. vllm/model_executor/layers/rotary_embedding/ernie45_vl_rope.py +80 -0
  917. vllm/model_executor/layers/rotary_embedding/linear_scaling_rope.py +115 -0
  918. vllm/model_executor/layers/rotary_embedding/llama3_rope.py +54 -0
  919. vllm/model_executor/layers/rotary_embedding/llama4_vision_rope.py +81 -0
  920. vllm/model_executor/layers/rotary_embedding/mrope.py +1321 -0
  921. vllm/model_executor/layers/rotary_embedding/ntk_scaling_rope.py +42 -0
  922. vllm/model_executor/layers/rotary_embedding/phi3_long_rope_scaled_rope.py +129 -0
  923. vllm/model_executor/layers/rotary_embedding/rocm_aiter_rope_ops.py +86 -0
  924. vllm/model_executor/layers/rotary_embedding/yarn_scaling_rope.py +68 -0
  925. vllm/model_executor/layers/shared_fused_moe/__init__.py +6 -0
  926. vllm/model_executor/layers/shared_fused_moe/shared_fused_moe.py +56 -0
  927. vllm/model_executor/layers/utils.py +195 -0
  928. vllm/model_executor/layers/vocab_parallel_embedding.py +487 -0
  929. vllm/model_executor/model_loader/__init__.py +138 -0
  930. vllm/model_executor/model_loader/base_loader.py +52 -0
  931. vllm/model_executor/model_loader/bitsandbytes_loader.py +788 -0
  932. vllm/model_executor/model_loader/default_loader.py +277 -0
  933. vllm/model_executor/model_loader/dummy_loader.py +28 -0
  934. vllm/model_executor/model_loader/gguf_loader.py +155 -0
  935. vllm/model_executor/model_loader/runai_streamer_loader.py +104 -0
  936. vllm/model_executor/model_loader/sharded_state_loader.py +199 -0
  937. vllm/model_executor/model_loader/tensorizer.py +738 -0
  938. vllm/model_executor/model_loader/tensorizer_loader.py +143 -0
  939. vllm/model_executor/model_loader/tpu.py +114 -0
  940. vllm/model_executor/model_loader/utils.py +292 -0
  941. vllm/model_executor/model_loader/weight_utils.py +990 -0
  942. vllm/model_executor/models/__init__.py +33 -0
  943. vllm/model_executor/models/adapters.py +542 -0
  944. vllm/model_executor/models/aimv2.py +246 -0
  945. vllm/model_executor/models/apertus.py +579 -0
  946. vllm/model_executor/models/arcee.py +422 -0
  947. vllm/model_executor/models/arctic.py +558 -0
  948. vllm/model_executor/models/aria.py +650 -0
  949. vllm/model_executor/models/aya_vision.py +468 -0
  950. vllm/model_executor/models/baichuan.py +474 -0
  951. vllm/model_executor/models/bailing_moe.py +642 -0
  952. vllm/model_executor/models/bamba.py +514 -0
  953. vllm/model_executor/models/bert.py +665 -0
  954. vllm/model_executor/models/bert_with_rope.py +687 -0
  955. vllm/model_executor/models/blip.py +339 -0
  956. vllm/model_executor/models/blip2.py +712 -0
  957. vllm/model_executor/models/bloom.py +374 -0
  958. vllm/model_executor/models/chameleon.py +1139 -0
  959. vllm/model_executor/models/chatglm.py +476 -0
  960. vllm/model_executor/models/clip.py +407 -0
  961. vllm/model_executor/models/cohere2_vision.py +481 -0
  962. vllm/model_executor/models/commandr.py +465 -0
  963. vllm/model_executor/models/config.py +445 -0
  964. vllm/model_executor/models/dbrx.py +471 -0
  965. vllm/model_executor/models/deepseek.py +497 -0
  966. vllm/model_executor/models/deepseek_eagle.py +240 -0
  967. vllm/model_executor/models/deepseek_mtp.py +289 -0
  968. vllm/model_executor/models/deepseek_v2.py +1444 -0
  969. vllm/model_executor/models/deepseek_vl2.py +658 -0
  970. vllm/model_executor/models/dots1.py +546 -0
  971. vllm/model_executor/models/dots_ocr.py +873 -0
  972. vllm/model_executor/models/ernie45.py +43 -0
  973. vllm/model_executor/models/ernie45_moe.py +607 -0
  974. vllm/model_executor/models/ernie45_vl.py +1527 -0
  975. vllm/model_executor/models/ernie45_vl_moe.py +727 -0
  976. vllm/model_executor/models/ernie_mtp.py +268 -0
  977. vllm/model_executor/models/exaone.py +550 -0
  978. vllm/model_executor/models/exaone4.py +533 -0
  979. vllm/model_executor/models/fairseq2_llama.py +154 -0
  980. vllm/model_executor/models/falcon.py +509 -0
  981. vllm/model_executor/models/falcon_h1.py +674 -0
  982. vllm/model_executor/models/fuyu.py +399 -0
  983. vllm/model_executor/models/gemma.py +425 -0
  984. vllm/model_executor/models/gemma2.py +422 -0
  985. vllm/model_executor/models/gemma3.py +555 -0
  986. vllm/model_executor/models/gemma3_mm.py +721 -0
  987. vllm/model_executor/models/gemma3n.py +1113 -0
  988. vllm/model_executor/models/gemma3n_mm.py +761 -0
  989. vllm/model_executor/models/glm.py +23 -0
  990. vllm/model_executor/models/glm4.py +304 -0
  991. vllm/model_executor/models/glm4_1v.py +1690 -0
  992. vllm/model_executor/models/glm4_moe.py +727 -0
  993. vllm/model_executor/models/glm4_moe_mtp.py +301 -0
  994. vllm/model_executor/models/glm4v.py +654 -0
  995. vllm/model_executor/models/gpt2.py +380 -0
  996. vllm/model_executor/models/gpt_bigcode.py +344 -0
  997. vllm/model_executor/models/gpt_j.py +339 -0
  998. vllm/model_executor/models/gpt_neox.py +330 -0
  999. vllm/model_executor/models/gpt_oss.py +712 -0
  1000. vllm/model_executor/models/granite.py +489 -0
  1001. vllm/model_executor/models/granite_speech.py +794 -0
  1002. vllm/model_executor/models/granitemoe.py +550 -0
  1003. vllm/model_executor/models/granitemoehybrid.py +614 -0
  1004. vllm/model_executor/models/granitemoeshared.py +332 -0
  1005. vllm/model_executor/models/gritlm.py +262 -0
  1006. vllm/model_executor/models/grok1.py +547 -0
  1007. vllm/model_executor/models/h2ovl.py +536 -0
  1008. vllm/model_executor/models/hunyuan_v1.py +1042 -0
  1009. vllm/model_executor/models/hyperclovax_vision.py +1192 -0
  1010. vllm/model_executor/models/idefics2_vision_model.py +417 -0
  1011. vllm/model_executor/models/idefics3.py +756 -0
  1012. vllm/model_executor/models/interfaces.py +959 -0
  1013. vllm/model_executor/models/interfaces_base.py +192 -0
  1014. vllm/model_executor/models/intern_vit.py +441 -0
  1015. vllm/model_executor/models/internlm2.py +450 -0
  1016. vllm/model_executor/models/internlm2_ve.py +148 -0
  1017. vllm/model_executor/models/interns1.py +838 -0
  1018. vllm/model_executor/models/interns1_vit.py +418 -0
  1019. vllm/model_executor/models/internvl.py +1423 -0
  1020. vllm/model_executor/models/jais.py +373 -0
  1021. vllm/model_executor/models/jamba.py +591 -0
  1022. vllm/model_executor/models/jina_vl.py +144 -0
  1023. vllm/model_executor/models/keye.py +1680 -0
  1024. vllm/model_executor/models/keye_vl1_5.py +602 -0
  1025. vllm/model_executor/models/kimi_vl.py +618 -0
  1026. vllm/model_executor/models/lfm2.py +548 -0
  1027. vllm/model_executor/models/llama.py +669 -0
  1028. vllm/model_executor/models/llama4.py +746 -0
  1029. vllm/model_executor/models/llama4_eagle.py +239 -0
  1030. vllm/model_executor/models/llama_eagle.py +179 -0
  1031. vllm/model_executor/models/llama_eagle3.py +296 -0
  1032. vllm/model_executor/models/llava.py +870 -0
  1033. vllm/model_executor/models/llava_next.py +571 -0
  1034. vllm/model_executor/models/llava_next_video.py +476 -0
  1035. vllm/model_executor/models/llava_onevision.py +942 -0
  1036. vllm/model_executor/models/longcat_flash.py +715 -0
  1037. vllm/model_executor/models/longcat_flash_mtp.py +352 -0
  1038. vllm/model_executor/models/mamba.py +275 -0
  1039. vllm/model_executor/models/mamba2.py +291 -0
  1040. vllm/model_executor/models/medusa.py +169 -0
  1041. vllm/model_executor/models/midashenglm.py +792 -0
  1042. vllm/model_executor/models/mimo.py +188 -0
  1043. vllm/model_executor/models/mimo_mtp.py +280 -0
  1044. vllm/model_executor/models/minicpm.py +631 -0
  1045. vllm/model_executor/models/minicpm3.py +230 -0
  1046. vllm/model_executor/models/minicpm_eagle.py +389 -0
  1047. vllm/model_executor/models/minicpmo.py +770 -0
  1048. vllm/model_executor/models/minicpmv.py +1784 -0
  1049. vllm/model_executor/models/minimax_text_01.py +986 -0
  1050. vllm/model_executor/models/minimax_vl_01.py +426 -0
  1051. vllm/model_executor/models/mistral3.py +628 -0
  1052. vllm/model_executor/models/mixtral.py +606 -0
  1053. vllm/model_executor/models/mllama4.py +1076 -0
  1054. vllm/model_executor/models/mlp_speculator.py +206 -0
  1055. vllm/model_executor/models/modernbert.py +374 -0
  1056. vllm/model_executor/models/module_mapping.py +72 -0
  1057. vllm/model_executor/models/molmo.py +1567 -0
  1058. vllm/model_executor/models/moonvit.py +673 -0
  1059. vllm/model_executor/models/motif.py +345 -0
  1060. vllm/model_executor/models/mpt.py +329 -0
  1061. vllm/model_executor/models/nano_nemotron_vl.py +1394 -0
  1062. vllm/model_executor/models/nemotron.py +507 -0
  1063. vllm/model_executor/models/nemotron_h.py +565 -0
  1064. vllm/model_executor/models/nemotron_nas.py +481 -0
  1065. vllm/model_executor/models/nemotron_vl.py +652 -0
  1066. vllm/model_executor/models/nvlm_d.py +203 -0
  1067. vllm/model_executor/models/olmo.py +404 -0
  1068. vllm/model_executor/models/olmo2.py +439 -0
  1069. vllm/model_executor/models/olmoe.py +483 -0
  1070. vllm/model_executor/models/opt.py +412 -0
  1071. vllm/model_executor/models/orion.py +348 -0
  1072. vllm/model_executor/models/ovis.py +559 -0
  1073. vllm/model_executor/models/ovis2_5.py +642 -0
  1074. vllm/model_executor/models/paligemma.py +411 -0
  1075. vllm/model_executor/models/persimmon.py +343 -0
  1076. vllm/model_executor/models/phi.py +356 -0
  1077. vllm/model_executor/models/phi3.py +19 -0
  1078. vllm/model_executor/models/phi3v.py +698 -0
  1079. vllm/model_executor/models/phi4_multimodal.py +1475 -0
  1080. vllm/model_executor/models/phi4mm.py +1279 -0
  1081. vllm/model_executor/models/phi4mm_audio.py +1254 -0
  1082. vllm/model_executor/models/phi4mm_utils.py +1875 -0
  1083. vllm/model_executor/models/phimoe.py +679 -0
  1084. vllm/model_executor/models/pixtral.py +1345 -0
  1085. vllm/model_executor/models/plamo2.py +978 -0
  1086. vllm/model_executor/models/qwen.py +361 -0
  1087. vllm/model_executor/models/qwen2.py +523 -0
  1088. vllm/model_executor/models/qwen2_5_omni_thinker.py +984 -0
  1089. vllm/model_executor/models/qwen2_5_vl.py +1481 -0
  1090. vllm/model_executor/models/qwen2_audio.py +489 -0
  1091. vllm/model_executor/models/qwen2_moe.py +558 -0
  1092. vllm/model_executor/models/qwen2_rm.py +122 -0
  1093. vllm/model_executor/models/qwen2_vl.py +1670 -0
  1094. vllm/model_executor/models/qwen3.py +341 -0
  1095. vllm/model_executor/models/qwen3_moe.py +692 -0
  1096. vllm/model_executor/models/qwen3_next.py +1266 -0
  1097. vllm/model_executor/models/qwen3_next_mtp.py +281 -0
  1098. vllm/model_executor/models/qwen3_vl.py +1613 -0
  1099. vllm/model_executor/models/qwen3_vl_moe.py +358 -0
  1100. vllm/model_executor/models/qwen_vl.py +795 -0
  1101. vllm/model_executor/models/radio.py +576 -0
  1102. vllm/model_executor/models/registry.py +990 -0
  1103. vllm/model_executor/models/roberta.py +252 -0
  1104. vllm/model_executor/models/rvl.py +103 -0
  1105. vllm/model_executor/models/seed_oss.py +485 -0
  1106. vllm/model_executor/models/siglip.py +540 -0
  1107. vllm/model_executor/models/siglip2navit.py +689 -0
  1108. vllm/model_executor/models/skyworkr1v.py +911 -0
  1109. vllm/model_executor/models/smolvlm.py +44 -0
  1110. vllm/model_executor/models/solar.py +504 -0
  1111. vllm/model_executor/models/stablelm.py +341 -0
  1112. vllm/model_executor/models/starcoder2.py +354 -0
  1113. vllm/model_executor/models/step3_text.py +510 -0
  1114. vllm/model_executor/models/step3_vl.py +1072 -0
  1115. vllm/model_executor/models/swin.py +475 -0
  1116. vllm/model_executor/models/tarsier.py +639 -0
  1117. vllm/model_executor/models/telechat2.py +151 -0
  1118. vllm/model_executor/models/teleflm.py +79 -0
  1119. vllm/model_executor/models/terratorch.py +294 -0
  1120. vllm/model_executor/models/transformers.py +948 -0
  1121. vllm/model_executor/models/ultravox.py +654 -0
  1122. vllm/model_executor/models/utils.py +808 -0
  1123. vllm/model_executor/models/vision.py +404 -0
  1124. vllm/model_executor/models/voxtral.py +786 -0
  1125. vllm/model_executor/models/whisper.py +963 -0
  1126. vllm/model_executor/models/zamba2.py +960 -0
  1127. vllm/model_executor/parameter.py +620 -0
  1128. vllm/model_executor/utils.py +86 -0
  1129. vllm/model_executor/warmup/__init__.py +0 -0
  1130. vllm/model_executor/warmup/deep_gemm_warmup.py +230 -0
  1131. vllm/model_executor/warmup/kernel_warmup.py +83 -0
  1132. vllm/multimodal/__init__.py +33 -0
  1133. vllm/multimodal/audio.py +116 -0
  1134. vllm/multimodal/base.py +27 -0
  1135. vllm/multimodal/cache.py +697 -0
  1136. vllm/multimodal/evs.py +273 -0
  1137. vllm/multimodal/hasher.py +102 -0
  1138. vllm/multimodal/image.py +130 -0
  1139. vllm/multimodal/inputs.py +987 -0
  1140. vllm/multimodal/parse.py +511 -0
  1141. vllm/multimodal/processing.py +2148 -0
  1142. vllm/multimodal/profiling.py +284 -0
  1143. vllm/multimodal/registry.py +345 -0
  1144. vllm/multimodal/utils.py +503 -0
  1145. vllm/multimodal/video.py +319 -0
  1146. vllm/outputs.py +324 -0
  1147. vllm/platforms/__init__.py +263 -0
  1148. vllm/platforms/cpu.py +340 -0
  1149. vllm/platforms/cuda.py +668 -0
  1150. vllm/platforms/interface.py +620 -0
  1151. vllm/platforms/rocm.py +497 -0
  1152. vllm/platforms/tpu.py +233 -0
  1153. vllm/platforms/xpu.py +243 -0
  1154. vllm/plugins/__init__.py +72 -0
  1155. vllm/plugins/io_processors/__init__.py +68 -0
  1156. vllm/plugins/io_processors/interface.py +67 -0
  1157. vllm/plugins/lora_resolvers/README.md +16 -0
  1158. vllm/plugins/lora_resolvers/__init__.py +0 -0
  1159. vllm/plugins/lora_resolvers/filesystem_resolver.py +50 -0
  1160. vllm/pooling_params.py +191 -0
  1161. vllm/profiler/__init__.py +0 -0
  1162. vllm/profiler/layerwise_profile.py +375 -0
  1163. vllm/profiler/utils.py +148 -0
  1164. vllm/py.typed +2 -0
  1165. vllm/ray/__init__.py +0 -0
  1166. vllm/ray/lazy_utils.py +22 -0
  1167. vllm/ray/ray_env.py +72 -0
  1168. vllm/reasoning/__init__.py +29 -0
  1169. vllm/reasoning/abs_reasoning_parsers.py +202 -0
  1170. vllm/reasoning/basic_parsers.py +156 -0
  1171. vllm/reasoning/deepseek_r1_reasoning_parser.py +67 -0
  1172. vllm/reasoning/glm4_moe_reasoning_parser.py +151 -0
  1173. vllm/reasoning/gptoss_reasoning_parser.py +87 -0
  1174. vllm/reasoning/granite_reasoning_parser.py +363 -0
  1175. vllm/reasoning/hunyuan_a13b_reasoning_parser.py +245 -0
  1176. vllm/reasoning/mistral_reasoning_parser.py +56 -0
  1177. vllm/reasoning/qwen3_reasoning_parser.py +72 -0
  1178. vllm/reasoning/seedoss_reasoning_parser.py +28 -0
  1179. vllm/reasoning/step3_reasoning_parser.py +109 -0
  1180. vllm/sampling_params.py +593 -0
  1181. vllm/scalar_type.py +349 -0
  1182. vllm/scripts.py +15 -0
  1183. vllm/sequence.py +103 -0
  1184. vllm/tasks.py +11 -0
  1185. vllm/test_utils.py +129 -0
  1186. vllm/third_party/__init__.py +0 -0
  1187. vllm/third_party/pynvml.py +6140 -0
  1188. vllm/tracing.py +136 -0
  1189. vllm/transformers_utils/__init__.py +24 -0
  1190. vllm/transformers_utils/chat_templates/__init__.py +5 -0
  1191. vllm/transformers_utils/chat_templates/registry.py +70 -0
  1192. vllm/transformers_utils/chat_templates/template_basic.jinja +3 -0
  1193. vllm/transformers_utils/chat_templates/template_blip2.jinja +11 -0
  1194. vllm/transformers_utils/chat_templates/template_chatml.jinja +10 -0
  1195. vllm/transformers_utils/chat_templates/template_deepseek_vl2.jinja +23 -0
  1196. vllm/transformers_utils/chat_templates/template_fuyu.jinja +3 -0
  1197. vllm/transformers_utils/chat_templates/template_minicpmv45.jinja +93 -0
  1198. vllm/transformers_utils/config.py +1102 -0
  1199. vllm/transformers_utils/config_parser_base.py +20 -0
  1200. vllm/transformers_utils/configs/__init__.py +63 -0
  1201. vllm/transformers_utils/configs/arctic.py +207 -0
  1202. vllm/transformers_utils/configs/chatglm.py +72 -0
  1203. vllm/transformers_utils/configs/deepseek_v3.py +101 -0
  1204. vllm/transformers_utils/configs/deepseek_vl2.py +216 -0
  1205. vllm/transformers_utils/configs/dotsocr.py +69 -0
  1206. vllm/transformers_utils/configs/eagle.py +84 -0
  1207. vllm/transformers_utils/configs/falcon.py +90 -0
  1208. vllm/transformers_utils/configs/jais.py +237 -0
  1209. vllm/transformers_utils/configs/kimi_vl.py +37 -0
  1210. vllm/transformers_utils/configs/medusa.py +63 -0
  1211. vllm/transformers_utils/configs/midashenglm.py +101 -0
  1212. vllm/transformers_utils/configs/mistral.py +165 -0
  1213. vllm/transformers_utils/configs/mlp_speculator.py +68 -0
  1214. vllm/transformers_utils/configs/moonvit.py +33 -0
  1215. vllm/transformers_utils/configs/nemotron.py +205 -0
  1216. vllm/transformers_utils/configs/nemotron_h.py +259 -0
  1217. vllm/transformers_utils/configs/nemotron_vl.py +56 -0
  1218. vllm/transformers_utils/configs/olmo3.py +80 -0
  1219. vllm/transformers_utils/configs/ovis.py +176 -0
  1220. vllm/transformers_utils/configs/qwen3_next.py +275 -0
  1221. vllm/transformers_utils/configs/radio.py +91 -0
  1222. vllm/transformers_utils/configs/speculators/__init__.py +2 -0
  1223. vllm/transformers_utils/configs/speculators/algos.py +32 -0
  1224. vllm/transformers_utils/configs/speculators/base.py +111 -0
  1225. vllm/transformers_utils/configs/step3_vl.py +123 -0
  1226. vllm/transformers_utils/configs/ultravox.py +116 -0
  1227. vllm/transformers_utils/detokenizer_utils.py +199 -0
  1228. vllm/transformers_utils/dynamic_module.py +60 -0
  1229. vllm/transformers_utils/processor.py +299 -0
  1230. vllm/transformers_utils/processors/__init__.py +16 -0
  1231. vllm/transformers_utils/processors/deepseek_vl2.py +362 -0
  1232. vllm/transformers_utils/processors/ovis.py +420 -0
  1233. vllm/transformers_utils/processors/ovis2_5.py +458 -0
  1234. vllm/transformers_utils/runai_utils.py +104 -0
  1235. vllm/transformers_utils/s3_utils.py +93 -0
  1236. vllm/transformers_utils/tokenizer.py +292 -0
  1237. vllm/transformers_utils/tokenizer_base.py +154 -0
  1238. vllm/transformers_utils/tokenizers/__init__.py +10 -0
  1239. vllm/transformers_utils/tokenizers/mistral.py +521 -0
  1240. vllm/transformers_utils/utils.py +108 -0
  1241. vllm/triton_utils/__init__.py +16 -0
  1242. vllm/triton_utils/importing.py +96 -0
  1243. vllm/usage/__init__.py +0 -0
  1244. vllm/usage/usage_lib.py +259 -0
  1245. vllm/utils/__init__.py +3566 -0
  1246. vllm/utils/deep_gemm.py +319 -0
  1247. vllm/utils/flashinfer.py +443 -0
  1248. vllm/utils/jsontree.py +178 -0
  1249. vllm/utils/tensor_schema.py +235 -0
  1250. vllm/v1/__init__.py +0 -0
  1251. vllm/v1/attention/__init__.py +0 -0
  1252. vllm/v1/attention/backends/__init__.py +0 -0
  1253. vllm/v1/attention/backends/cpu_attn.py +919 -0
  1254. vllm/v1/attention/backends/flash_attn.py +795 -0
  1255. vllm/v1/attention/backends/flashinfer.py +1181 -0
  1256. vllm/v1/attention/backends/flex_attention.py +861 -0
  1257. vllm/v1/attention/backends/gdn_attn.py +332 -0
  1258. vllm/v1/attention/backends/linear_attn.py +67 -0
  1259. vllm/v1/attention/backends/mamba1_attn.py +81 -0
  1260. vllm/v1/attention/backends/mamba2_attn.py +232 -0
  1261. vllm/v1/attention/backends/mamba_attn.py +52 -0
  1262. vllm/v1/attention/backends/mla/__init__.py +0 -0
  1263. vllm/v1/attention/backends/mla/common.py +1783 -0
  1264. vllm/v1/attention/backends/mla/cutlass_mla.py +248 -0
  1265. vllm/v1/attention/backends/mla/flashattn_mla.py +271 -0
  1266. vllm/v1/attention/backends/mla/flashinfer_mla.py +114 -0
  1267. vllm/v1/attention/backends/mla/flashmla.py +203 -0
  1268. vllm/v1/attention/backends/mla/flashmla_sparse.py +544 -0
  1269. vllm/v1/attention/backends/mla/indexer.py +342 -0
  1270. vllm/v1/attention/backends/mla/rocm_aiter_mla.py +255 -0
  1271. vllm/v1/attention/backends/mla/triton_mla.py +177 -0
  1272. vllm/v1/attention/backends/pallas.py +409 -0
  1273. vllm/v1/attention/backends/rocm_aiter_fa.py +549 -0
  1274. vllm/v1/attention/backends/rocm_attn.py +426 -0
  1275. vllm/v1/attention/backends/short_conv_attn.py +94 -0
  1276. vllm/v1/attention/backends/tree_attn.py +451 -0
  1277. vllm/v1/attention/backends/triton_attn.py +361 -0
  1278. vllm/v1/attention/backends/utils.py +990 -0
  1279. vllm/v1/attention/backends/xformers.py +438 -0
  1280. vllm/v1/core/__init__.py +0 -0
  1281. vllm/v1/core/block_pool.py +416 -0
  1282. vllm/v1/core/encoder_cache_manager.py +333 -0
  1283. vllm/v1/core/kv_cache_coordinator.py +440 -0
  1284. vllm/v1/core/kv_cache_manager.py +399 -0
  1285. vllm/v1/core/kv_cache_utils.py +1291 -0
  1286. vllm/v1/core/sched/__init__.py +0 -0
  1287. vllm/v1/core/sched/async_scheduler.py +47 -0
  1288. vllm/v1/core/sched/interface.py +158 -0
  1289. vllm/v1/core/sched/output.py +166 -0
  1290. vllm/v1/core/sched/request_queue.py +224 -0
  1291. vllm/v1/core/sched/scheduler.py +1296 -0
  1292. vllm/v1/core/sched/utils.py +69 -0
  1293. vllm/v1/core/single_type_kv_cache_manager.py +671 -0
  1294. vllm/v1/cudagraph_dispatcher.py +125 -0
  1295. vllm/v1/engine/__init__.py +203 -0
  1296. vllm/v1/engine/async_llm.py +742 -0
  1297. vllm/v1/engine/coordinator.py +357 -0
  1298. vllm/v1/engine/core.py +1235 -0
  1299. vllm/v1/engine/core_client.py +1334 -0
  1300. vllm/v1/engine/detokenizer.py +349 -0
  1301. vllm/v1/engine/exceptions.py +17 -0
  1302. vllm/v1/engine/llm_engine.py +370 -0
  1303. vllm/v1/engine/logprobs.py +201 -0
  1304. vllm/v1/engine/output_processor.py +576 -0
  1305. vllm/v1/engine/parallel_sampling.py +133 -0
  1306. vllm/v1/engine/processor.py +545 -0
  1307. vllm/v1/engine/utils.py +860 -0
  1308. vllm/v1/executor/__init__.py +0 -0
  1309. vllm/v1/executor/abstract.py +137 -0
  1310. vllm/v1/executor/multiproc_executor.py +726 -0
  1311. vllm/v1/executor/ray_distributed_executor.py +108 -0
  1312. vllm/v1/executor/utils.py +23 -0
  1313. vllm/v1/kv_cache_interface.py +375 -0
  1314. vllm/v1/kv_offload/__init__.py +0 -0
  1315. vllm/v1/kv_offload/abstract.py +165 -0
  1316. vllm/v1/kv_offload/backend.py +96 -0
  1317. vllm/v1/kv_offload/backends/__init__.py +0 -0
  1318. vllm/v1/kv_offload/backends/cpu.py +61 -0
  1319. vllm/v1/kv_offload/cpu.py +75 -0
  1320. vllm/v1/kv_offload/factory.py +56 -0
  1321. vllm/v1/kv_offload/lru_manager.py +132 -0
  1322. vllm/v1/kv_offload/mediums.py +39 -0
  1323. vllm/v1/kv_offload/spec.py +61 -0
  1324. vllm/v1/kv_offload/worker/__init__.py +0 -0
  1325. vllm/v1/kv_offload/worker/cpu_gpu.py +171 -0
  1326. vllm/v1/kv_offload/worker/worker.py +142 -0
  1327. vllm/v1/metrics/__init__.py +0 -0
  1328. vllm/v1/metrics/loggers.py +741 -0
  1329. vllm/v1/metrics/prometheus.py +82 -0
  1330. vllm/v1/metrics/ray_wrappers.py +152 -0
  1331. vllm/v1/metrics/reader.py +246 -0
  1332. vllm/v1/metrics/stats.py +257 -0
  1333. vllm/v1/outputs.py +161 -0
  1334. vllm/v1/pool/__init__.py +0 -0
  1335. vllm/v1/pool/metadata.py +77 -0
  1336. vllm/v1/request.py +241 -0
  1337. vllm/v1/sample/__init__.py +0 -0
  1338. vllm/v1/sample/logits_processor/__init__.py +294 -0
  1339. vllm/v1/sample/logits_processor/builtin.py +275 -0
  1340. vllm/v1/sample/logits_processor/interface.py +97 -0
  1341. vllm/v1/sample/logits_processor/state.py +161 -0
  1342. vllm/v1/sample/metadata.py +43 -0
  1343. vllm/v1/sample/ops/__init__.py +0 -0
  1344. vllm/v1/sample/ops/bad_words.py +39 -0
  1345. vllm/v1/sample/ops/logprobs.py +26 -0
  1346. vllm/v1/sample/ops/penalties.py +43 -0
  1347. vllm/v1/sample/ops/topk_topp_sampler.py +292 -0
  1348. vllm/v1/sample/rejection_sampler.py +623 -0
  1349. vllm/v1/sample/sampler.py +285 -0
  1350. vllm/v1/sample/tpu/__init__.py +0 -0
  1351. vllm/v1/sample/tpu/metadata.py +124 -0
  1352. vllm/v1/sample/tpu/sampler.py +213 -0
  1353. vllm/v1/serial_utils.py +423 -0
  1354. vllm/v1/spec_decode/__init__.py +0 -0
  1355. vllm/v1/spec_decode/eagle.py +1011 -0
  1356. vllm/v1/spec_decode/medusa.py +66 -0
  1357. vllm/v1/spec_decode/metadata.py +62 -0
  1358. vllm/v1/spec_decode/metrics.py +211 -0
  1359. vllm/v1/spec_decode/ngram_proposer.py +276 -0
  1360. vllm/v1/spec_decode/utils.py +14 -0
  1361. vllm/v1/structured_output/__init__.py +295 -0
  1362. vllm/v1/structured_output/backend_guidance.py +245 -0
  1363. vllm/v1/structured_output/backend_lm_format_enforcer.py +167 -0
  1364. vllm/v1/structured_output/backend_outlines.py +320 -0
  1365. vllm/v1/structured_output/backend_types.py +134 -0
  1366. vllm/v1/structured_output/backend_xgrammar.py +327 -0
  1367. vllm/v1/structured_output/request.py +86 -0
  1368. vllm/v1/structured_output/utils.py +454 -0
  1369. vllm/v1/utils.py +396 -0
  1370. vllm/v1/worker/__init__.py +0 -0
  1371. vllm/v1/worker/block_table.py +210 -0
  1372. vllm/v1/worker/cpu_model_runner.py +175 -0
  1373. vllm/v1/worker/cpu_worker.py +156 -0
  1374. vllm/v1/worker/gpu_input_batch.py +863 -0
  1375. vllm/v1/worker/gpu_model_runner.py +4160 -0
  1376. vllm/v1/worker/gpu_ubatch_wrapper.py +399 -0
  1377. vllm/v1/worker/gpu_worker.py +710 -0
  1378. vllm/v1/worker/kv_connector_model_runner_mixin.py +132 -0
  1379. vllm/v1/worker/lora_model_runner_mixin.py +183 -0
  1380. vllm/v1/worker/tpu_input_batch.py +587 -0
  1381. vllm/v1/worker/tpu_model_runner.py +1946 -0
  1382. vllm/v1/worker/tpu_worker.py +346 -0
  1383. vllm/v1/worker/ubatch_splitting.py +192 -0
  1384. vllm/v1/worker/ubatch_utils.py +27 -0
  1385. vllm/v1/worker/ubatching.py +224 -0
  1386. vllm/v1/worker/utils.py +344 -0
  1387. vllm/v1/worker/worker_base.py +65 -0
  1388. vllm/v1/worker/xpu_model_runner.py +57 -0
  1389. vllm/v1/worker/xpu_worker.py +179 -0
  1390. vllm/version.py +41 -0
  1391. vllm/vllm_flash_attn/.gitkeep +0 -0
  1392. vllm/worker/__init__.py +0 -0
  1393. vllm/worker/worker_base.py +279 -0
  1394. vllm_cpu-0.11.0.post2.dist-info/METADATA +348 -0
  1395. vllm_cpu-0.11.0.post2.dist-info/RECORD +1398 -0
  1396. vllm_cpu-0.11.0.post2.dist-info/WHEEL +5 -0
  1397. vllm_cpu-0.11.0.post2.dist-info/entry_points.txt +5 -0
  1398. vllm_cpu-0.11.0.post2.dist-info/top_level.txt +1 -0
@@ -0,0 +1,1137 @@
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, Optional, Union
8
+ from xml.parsers.expat import ParserCreate
9
+
10
+ import regex as re
11
+
12
+ from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
13
+ ChatCompletionToolsParam,
14
+ DeltaFunctionCall, DeltaMessage,
15
+ DeltaToolCall,
16
+ ExtractedToolCallInformation,
17
+ FunctionCall, ToolCall)
18
+ from vllm.entrypoints.openai.tool_parsers.abstract_tool_parser import (
19
+ ToolParser, ToolParserManager)
20
+ from vllm.logger import init_logger
21
+ from vllm.transformers_utils.tokenizer import AnyTokenizer
22
+
23
+ logger = init_logger(__name__)
24
+
25
+
26
+ class StreamingXMLToolCallParser:
27
+ """
28
+ Simplified streaming XML tool call parser
29
+ Supports streaming input, parsing, and output
30
+ """
31
+
32
+ def __init__(self):
33
+ self.reset_streaming_state()
34
+
35
+ # Tool configuration information
36
+ self.tools: Union[list[ChatCompletionToolsParam], None] = None
37
+ self.tool_call_start_token: str = '<tool_call>'
38
+ self.tool_call_end_token: str = '</tool_call>'
39
+ self.function_start_token: str = '<function='
40
+ self.function_end_token: str = '</function>'
41
+ self.parameter_start_token: str = '<parameter='
42
+ self.parameter_end_token: str = '</parameter>'
43
+
44
+ def reset_streaming_state(self):
45
+ """Reset streaming parsing state"""
46
+
47
+ self.deltas = []
48
+ # state for streaming
49
+ self.tool_call_index = 0
50
+ self.current_call_id = None
51
+ self.last_completed_call_id = None
52
+ self.current_function_name = None
53
+ self.current_function_open = False
54
+ self.parameters = {}
55
+ self.current_param_name = None
56
+ self.current_param_value = ''
57
+ self.current_param_value_converted = ''
58
+ self.current_param_is_first = False
59
+ self.should_emit_end_newline = False
60
+ self.start_quote_emitted = False
61
+
62
+ self.streaming_buffer = ''
63
+ self.last_processed_pos = 0
64
+
65
+ self.text_content_buffer = ''
66
+
67
+ # state for preprocessing and deferred parsing
68
+ self._pre_inside_parameter = False
69
+ self._pre_param_buffer = ""
70
+ self._pre_current_param_name = None
71
+ self.defer_current_parameter = False
72
+ self.deferred_param_raw_value = ""
73
+
74
+ # recreate parser
75
+ self.parser = ParserCreate()
76
+ self.setup_parser()
77
+
78
+ def parse_single_streaming_chunks(self, xml_chunk: str) -> DeltaMessage:
79
+ """
80
+ Parse single streaming XML chunk and return Delta response
81
+ This is the actual streaming interface that receives chunks
82
+ one by one and maintains internal state
83
+
84
+ Args:
85
+ xml_chunk: Single XML chunk string
86
+ Returns:
87
+ DeltaMessage: Contains delta information generated by this chunk,
88
+ returns empty response if no complete elements
89
+ """
90
+ # Record delta count before processing
91
+ initial_delta_count = len(self.deltas)
92
+
93
+ self.streaming_buffer += xml_chunk
94
+
95
+ found_elements = self._process_complete_xml_elements()
96
+
97
+ if found_elements:
98
+ # If complete elements found, check if end events were missed
99
+ # some tags may not have been triggered
100
+ try:
101
+ new_deltas = self.deltas[initial_delta_count:]
102
+ # If this chunk contains </function>
103
+ # but didn't generate '}', then complete it
104
+ if (self.current_call_id is not None
105
+ and self.function_end_token in xml_chunk):
106
+
107
+ # - Added '}' (non-empty parameter ending)
108
+ # - Added '{}' (empty parameter function)
109
+ has_function_close = any((td.tool_calls and any(
110
+ (tc.function and tc.id == self.current_call_id
111
+ and isinstance(tc.function.arguments, str) and
112
+ (tc.function.arguments in ('}', '{}')))
113
+ for tc in td.tool_calls)) for td in new_deltas)
114
+ if not has_function_close:
115
+ # Close potentially unclosed element
116
+ if self.current_param_name:
117
+ self._end_element('parameter')
118
+ if self.current_function_name:
119
+ self._end_element('function')
120
+ # If this chunk contains </tool_call>
121
+ # but didn't generate final empty delta, then complete it
122
+ if (self.current_call_id is not None
123
+ and self.tool_call_end_token in xml_chunk):
124
+ has_toolcall_close = any((td.tool_calls and any(
125
+ (tc.type == 'function' and tc.function and tc.function.
126
+ arguments == '' and tc.id == self.current_call_id)
127
+ for tc in td.tool_calls)) for td in new_deltas)
128
+ if not has_toolcall_close:
129
+ # Close potentially unclosed element
130
+ if self.current_param_name:
131
+ self._end_element('parameter')
132
+ if self.current_function_name:
133
+ self._end_element('function')
134
+ self._end_element('tool_call')
135
+ except Exception as e:
136
+ logger.warning("Error with fallback parsing: %s", e)
137
+ # Merge newly generated deltas into single response
138
+ result_delta = self._merge_new_deltas_to_single_response(
139
+ initial_delta_count)
140
+ return result_delta
141
+ else:
142
+ # No complete elements, check if there's unoutput text content
143
+ if self.text_content_buffer and self.tool_call_index == 0:
144
+ # Has text content but no tool_call yet, output text content
145
+ text_delta = DeltaMessage(content=self.text_content_buffer)
146
+ self._emit_delta(text_delta)
147
+ # Clear buffer to avoid duplicate output
148
+ self.text_content_buffer = ''
149
+ return text_delta
150
+
151
+ # If this chunk contains end tags but wasn't triggered by parser,
152
+ # manually complete end events
153
+ # Only execute when still on the same call as when entered,
154
+ # to prevent accidentally closing new calls
155
+ # in multi <tool_call> scenarios
156
+ if (self.current_call_id is not None
157
+ and (self.function_end_token in xml_chunk
158
+ or self.tool_call_end_token in xml_chunk)):
159
+ # Close potentially unclosed element
160
+ if self.current_param_name:
161
+ self._end_element('parameter')
162
+ if self.function_end_token in xml_chunk and \
163
+ self.current_function_name:
164
+ self._end_element('function')
165
+ if self.tool_call_end_token in xml_chunk:
166
+ self._end_element('tool_call')
167
+ # Return the merged delta result generated by this fallback
168
+ result_delta = self._merge_new_deltas_to_single_response(
169
+ initial_delta_count)
170
+ return result_delta
171
+
172
+ # No complete elements, return empty response
173
+ return DeltaMessage(content=None)
174
+
175
+ def _escape_xml_special_chars(self, text: str) -> str:
176
+ """
177
+ Escape XML special characters
178
+ Args:
179
+ text: Original text
180
+ Returns:
181
+ Escaped text
182
+ """
183
+ xml_escapes = {
184
+ '&': '&amp;',
185
+ '<': '&lt;',
186
+ '>': '&gt;',
187
+ '"': '&quot;',
188
+ "'": '&apos;'
189
+ }
190
+
191
+ for char, escape in xml_escapes.items():
192
+ text = text.replace(char, escape)
193
+
194
+ return text
195
+
196
+ def _process_complete_xml_elements(self) -> bool:
197
+ """
198
+ Process complete XML elements in buffer
199
+
200
+ Returns:
201
+ bool: Whether complete elements were found and processed
202
+ """
203
+ found_any = False
204
+
205
+ while self.last_processed_pos < len(self.streaming_buffer):
206
+ # Find next complete xml element
207
+ element, end_pos = self._find_next_complete_element(
208
+ self.last_processed_pos)
209
+ if element is None:
210
+ # No complete element found, wait for more data
211
+ break
212
+
213
+ # Check if this element should be skipped
214
+ if self._should_skip_element(element):
215
+ self.last_processed_pos = end_pos
216
+ continue
217
+
218
+ # Found complete XML element, process it
219
+ try:
220
+ preprocessed_element = self._preprocess_xml_chunk(element)
221
+ # Check if this is the first tool_call start
222
+ if ((preprocessed_element.strip().startswith('<tool_call>') or
223
+ preprocessed_element.strip().startswith('<function name=')
224
+ ) and self.tool_call_index
225
+ == 0) and self.text_content_buffer:
226
+ # First tool_call starts,
227
+ # output previously collected text content first
228
+ text_delta = DeltaMessage(content=self.text_content_buffer)
229
+ self._emit_delta(text_delta)
230
+ # Clear buffer for potential subsequent text content
231
+ self.text_content_buffer = ''
232
+
233
+ # If a new tool_call starts and
234
+ # there are already completed tool_calls
235
+ if (preprocessed_element.strip().startswith('<tool_call>')
236
+ and self.tool_call_index > 0 and self.current_call_id):
237
+ # Reset parser state but preserve generated deltas
238
+ if self.current_param_name:
239
+ self._end_element('parameter')
240
+ if self.current_function_open or self.current_function_name:
241
+ self._end_element('function')
242
+ # Output final tool_call tail delta
243
+ final_delta = DeltaMessage(
244
+ role=None,
245
+ content=None,
246
+ reasoning_content=None,
247
+ tool_calls=[
248
+ DeltaToolCall(index=self.tool_call_index - 1,
249
+ id=self.current_call_id,
250
+ type='function',
251
+ function=DeltaFunctionCall(
252
+ name=None, arguments=''))
253
+ ])
254
+ self._emit_delta(final_delta)
255
+ # Reset XML parser and current call state
256
+ self._reset_xml_parser_after_tool_call()
257
+ # Parse preprocessed element
258
+ self.parser.Parse(preprocessed_element, False)
259
+ found_any = True
260
+
261
+ except Exception as e:
262
+ logger.warning("Error when parsing XML elements: %s", e)
263
+
264
+ # Update processed position
265
+ self.last_processed_pos = end_pos
266
+
267
+ return found_any
268
+
269
+ def _should_skip_element(self, element: str) -> bool:
270
+ """
271
+ Determine whether an element should be skipped
272
+
273
+ Args:
274
+ element: Element to evaluate
275
+
276
+ Returns:
277
+ bool: True means should skip, False means should process
278
+ """
279
+
280
+ # If it's a tool_call XML tag, don't skip
281
+ if element.startswith(
282
+ self.tool_call_start_token) or element.startswith(
283
+ self.function_start_token) or element.startswith(
284
+ self.parameter_start_token):
285
+ return False
286
+
287
+ # If currently not parsing tool calls and not blank,
288
+ # collect this text instead of skipping
289
+ # Only process other XML elements after tool_call appears,
290
+ # otherwise treat as plain text
291
+ if self.current_call_id is None and element:
292
+ # Collect text content to buffer
293
+ self.text_content_buffer += element
294
+ return True # Still skip, but content has been collected
295
+
296
+ # If currently parsing tool calls,
297
+ # this might be parameter value, don't skip
298
+ if self.current_call_id is not None:
299
+ return False
300
+
301
+ # Skip blank content
302
+ return not element
303
+
304
+ def _find_next_complete_element(
305
+ self, start_pos: int) -> tuple[Optional[str], int]:
306
+ """
307
+ Find next complete XML element from specified position
308
+
309
+ Args:
310
+ start_pos: Position to start searching
311
+
312
+ Returns:
313
+ (Complete element string, element end position),
314
+ returns (None, start_pos) if no complete element found
315
+ """
316
+ buffer = self.streaming_buffer[start_pos:]
317
+
318
+ if not buffer:
319
+ return None, start_pos
320
+
321
+ if buffer.startswith('<'):
322
+ # Need to ensure no new < appears,
323
+ # find the nearest one between < and >
324
+ tag_end = buffer.find('<', 1)
325
+ tag_end2 = buffer.find('>', 1)
326
+ if tag_end != -1 and tag_end2 != -1:
327
+ # Next nearest is <
328
+ if tag_end < tag_end2:
329
+ return buffer[:tag_end], start_pos + tag_end
330
+ # Next nearest is >, means found XML element
331
+ else:
332
+ return buffer[:tag_end2 + 1], start_pos + tag_end2 + 1
333
+ elif tag_end != -1:
334
+ return buffer[:tag_end], start_pos + tag_end
335
+ elif tag_end2 != -1:
336
+ return buffer[:tag_end2 + 1], start_pos + tag_end2 + 1
337
+ else:
338
+ # If currently not parsing tool calls (entering a tool_call),
339
+ # check if starts with <tool_call>
340
+ if self.current_call_id is None:
341
+ # Check if might be start of <tool_call>
342
+ if buffer == '<tool_call>'[:len(buffer)]:
343
+ # Might be start of <tool_call>, wait for more data
344
+ return None, start_pos
345
+ else:
346
+ # Not start of <tool_call>, treat as text
347
+ return buffer, start_pos + len(buffer)
348
+ else:
349
+ # When parsing tool calls,
350
+ # wait for more data to get complete tag
351
+ return None, start_pos
352
+ else:
353
+ # Find text content (until next < or buffer end)
354
+ next_tag_pos = buffer.find('<')
355
+ if next_tag_pos != -1:
356
+ # Found text content
357
+ text_content = buffer[:next_tag_pos]
358
+ return text_content, start_pos + next_tag_pos
359
+ else:
360
+ # Buffer end is all text, process
361
+ # (no longer wait for more data)
362
+ remaining = buffer
363
+ return remaining, start_pos + len(remaining)
364
+
365
+ def _merge_new_deltas_to_single_response(
366
+ self, initial_count: int) -> DeltaMessage:
367
+ """
368
+ Merge newly generated deltas from this processing
369
+ into a single DeltaMessage
370
+
371
+ Args:
372
+ initial_count: Delta count before processing
373
+
374
+ Returns:
375
+ Merged DeltaMessage containing all newly generated delta information
376
+ """
377
+ if len(self.deltas) <= initial_count:
378
+ return DeltaMessage(content=None)
379
+
380
+ # Get newly generated deltas
381
+ new_deltas = self.deltas[initial_count:]
382
+
383
+ if len(new_deltas) == 1:
384
+ # Only one new delta, return directly
385
+ return new_deltas[0]
386
+
387
+ # Merge multiple new deltas
388
+ merged_tool_calls: list[DeltaToolCall] = []
389
+ merged_content: str = ''
390
+
391
+ for delta in new_deltas:
392
+ if delta.content:
393
+ merged_content += delta.content
394
+ if delta.tool_calls:
395
+ # For tool_calls, we need to intelligently merge arguments
396
+ for tool_call in delta.tool_calls:
397
+ # Find if there's already a tool_call with the same call_id
398
+ existing_call = None
399
+ for existing in merged_tool_calls:
400
+ if existing.id == tool_call.id:
401
+ existing_call = existing
402
+ break
403
+
404
+ if existing_call and existing_call.function:
405
+ # Merge to existing tool_call
406
+ if tool_call.function and tool_call.function.name:
407
+ existing_call.function.name = \
408
+ tool_call.function.name
409
+ if tool_call.function \
410
+ and tool_call.function.arguments is not None:
411
+ if existing_call.function.arguments is None:
412
+ existing_call.function.arguments = ''
413
+
414
+ # For streaming JSON parameters,
415
+ # simply concatenate in order
416
+ new_args = tool_call.function.arguments
417
+ existing_call.function.arguments += new_args
418
+ if tool_call.type:
419
+ existing_call.type = tool_call.type
420
+ else:
421
+ # Add new tool_call
422
+ merged_tool_calls.append(tool_call)
423
+
424
+ return DeltaMessage(content=merged_content if merged_content else None,
425
+ tool_calls=merged_tool_calls)
426
+
427
+ def _preprocess_xml_chunk(self, chunk: str) -> str:
428
+ """
429
+ Preprocess XML chunk, handle non-standard formats,
430
+ and escape special characters
431
+
432
+ Args:
433
+ chunk: Original XML chunk
434
+
435
+ Returns:
436
+ Processed XML chunk
437
+ """
438
+
439
+ # Check if this is a tool_call related element
440
+ is_tool_call = False
441
+ if chunk.startswith(self.tool_call_start_token) or chunk.startswith(
442
+ self.tool_call_end_token):
443
+ is_tool_call = True
444
+ if chunk.startswith(self.function_start_token) or chunk.startswith(
445
+ self.function_end_token):
446
+ is_tool_call = True
447
+ if chunk.startswith(self.parameter_start_token) or chunk.startswith(
448
+ self.parameter_end_token):
449
+ is_tool_call = True
450
+ # Handle <function=name> format -> <function name="name">
451
+ processed = re.sub(r'<function=([^>]+)>', r'<function name="\1">',
452
+ chunk)
453
+ # Handle <parameter=name> format -> <parameter name="name">
454
+ processed = re.sub(r'<parameter=([^>]+)>', r'<parameter name="\1">',
455
+ processed)
456
+
457
+ original_chunk = chunk
458
+ # If in parameter value accumulation mode
459
+ if self._pre_inside_parameter:
460
+ # Parameter end: output accumulated raw text
461
+ # safely then return </parameter>
462
+ if processed.startswith('</parameter>'):
463
+ body_text = self._pre_param_buffer
464
+ # Trigger deferred parsing mode
465
+ # literal_eval+json output in end_element
466
+ self.defer_current_parameter = True
467
+ self.deferred_param_raw_value = body_text
468
+ # Clean up state
469
+ self._pre_inside_parameter = False
470
+ self._pre_param_buffer = ""
471
+ self._pre_current_param_name = None
472
+ safe_text = self._escape_xml_special_chars(body_text)
473
+ return f"{safe_text}</parameter>"
474
+ else:
475
+ # If this is the first block of content after entering parameter
476
+ # evaluate if deferred parsing is needed;
477
+ # If not needed, exit accumulation mode
478
+ # and pass through directly
479
+ if self._pre_param_buffer == "":
480
+ # Get current parameter type
481
+ param_type = self._get_param_type(
482
+ self._pre_current_param_name
483
+ ) if self._pre_current_param_name else 'string'
484
+ # Only these types need deferred parsing to
485
+ # handle Python literals containing single quotes
486
+ is_object_type = param_type in ["object"]
487
+ is_complex_type = (param_type
488
+ in ["array", "arr", "sequence"]
489
+ or param_type.startswith("dict")
490
+ or param_type.startswith("list"))
491
+
492
+ # Only delay when contains container symbols
493
+ # and has single quotes and is complex type
494
+ has_container_hint = ('[' in original_chunk) or (
495
+ '{' in original_chunk) or ('(' in original_chunk)
496
+
497
+ # Determine if deferred parsing is needed
498
+ need_defer = False
499
+ if is_complex_type:
500
+ # Complex type, always need deferred parsing
501
+ need_defer = True
502
+ elif is_object_type and has_container_hint and (
503
+ "'" in original_chunk):
504
+ # Object type with container symbols
505
+ # and single quotes, need deferred parsing
506
+ need_defer = True
507
+
508
+ if not need_defer:
509
+ # No need for deferred parsing,
510
+ # exit parameter mode directly
511
+ self._pre_inside_parameter = False
512
+ return self._escape_xml_special_chars(original_chunk)
513
+ self._pre_param_buffer += original_chunk
514
+ return ""
515
+
516
+ # Parameter start: enable accumulation
517
+ if processed.startswith('<parameter name='):
518
+ m = re.match(r'<parameter name="([^"]+)">', processed)
519
+ if m:
520
+ self._pre_current_param_name = m.group(1)
521
+ self._pre_inside_parameter = True
522
+ self._pre_param_buffer = ""
523
+ return processed
524
+
525
+ # If processed doesn't contain special_token, escape processed
526
+ # This is because XML parsing encounters special characters
527
+ # and reports errors, so escaping is needed
528
+ if not is_tool_call:
529
+ processed = self._escape_xml_special_chars(processed)
530
+ return processed
531
+
532
+ def _emit_delta(self, delta: DeltaMessage):
533
+ """Emit Delta response (streaming output)"""
534
+ self.deltas.append(delta)
535
+
536
+ def _auto_close_open_parameter_if_needed(self,
537
+ incoming_tag: Optional[str] = None
538
+ ):
539
+ """Before starting to process new elements,
540
+ if there are unclosed tags from before,
541
+ automatically complete their endings to the parser.
542
+ - If there are unclosed parameters,
543
+ it's equivalent to feeding `</parameter>`
544
+ - When about to start a new function or tool_call,
545
+ if there are unclosed functions, complete `</function>`.
546
+ - When about to start a new tool_call,
547
+ if there are unclosed tool_calls, complete `</tool_call>`.
548
+ """
549
+ # First close unclosed parameters
550
+ if self.current_param_name:
551
+ self._end_element('parameter')
552
+
553
+ # If about to start new function or tool_call,
554
+ # and there are unclosed functions, close function first
555
+ if incoming_tag in ('function',
556
+ 'tool_call') and self.current_function_name:
557
+ self._end_element('function')
558
+
559
+ # If about to start new tool_call,
560
+ # and there are unclosed tool_calls, close tool_call first
561
+ if incoming_tag == 'tool_call' and self.current_call_id:
562
+ self._end_element('tool_call')
563
+
564
+ def _start_element(self, name: str, attrs: dict[str, str]):
565
+ """Handle XML start element events"""
566
+
567
+ if name == 'root':
568
+ return
569
+
570
+ if name == 'tool_call':
571
+ # Before opening new tool_call,
572
+ # automatically complete previous unclosed tags
573
+ self._auto_close_open_parameter_if_needed('tool_call')
574
+
575
+ self.parameters = {}
576
+ self.current_call_id = self._get_next_call_id()
577
+ self.current_param_is_first = True
578
+ self.tool_call_index += 1
579
+ elif name.startswith('function') or (name == 'function'):
580
+ # If missing tool_call, manually complete
581
+ if not self.current_call_id:
582
+ self._start_element('tool_call', {})
583
+ # Before opening new function,
584
+ # automatically complete previous unclosed tags (parameter/function)
585
+ self._auto_close_open_parameter_if_needed('function')
586
+ function_name = self._extract_function_name(name, attrs)
587
+ self.current_function_name = function_name
588
+ self.current_function_open = True
589
+ if function_name:
590
+ delta = DeltaMessage(tool_calls=[
591
+ DeltaToolCall(index=self.tool_call_index - 1,
592
+ id=self.current_call_id,
593
+ type='function',
594
+ function=DeltaFunctionCall(
595
+ name=function_name, arguments=''))
596
+ ])
597
+ self._emit_delta(delta)
598
+ elif name.startswith('parameter') or (name == 'parameter'):
599
+ # If previous parameter hasn't ended normally,
600
+ # complete its end first, then start new parameter
601
+ self._auto_close_open_parameter_if_needed('parameter')
602
+ param_name = self._extract_parameter_name(name, attrs)
603
+ self.current_param_name = param_name
604
+ self.current_param_value = ''
605
+ self.current_param_value_converted = ''
606
+ self.start_quote_emitted = False # Reset start quote flag
607
+
608
+ # Only output parameter name and colon,
609
+ # don't output quotes
610
+ # decide after parameter value type is determined
611
+ if param_name:
612
+ if not self.parameters:
613
+ # First parameter
614
+ # start JSON, only output parameter name and colon
615
+ json_start = f'{{"{param_name}": '
616
+ delta = DeltaMessage(tool_calls=[
617
+ DeltaToolCall(index=self.tool_call_index - 1,
618
+ id=self.current_call_id,
619
+ type='function',
620
+ function=DeltaFunctionCall(
621
+ name=None, arguments=json_start))
622
+ ])
623
+ self._emit_delta(delta)
624
+ self.current_param_is_first = True
625
+ else:
626
+ # Subsequent parameters
627
+ # add comma and parameter name, no quotes
628
+ json_continue = f', "{param_name}": '
629
+ delta = DeltaMessage(tool_calls=[
630
+ DeltaToolCall(index=self.tool_call_index - 1,
631
+ id=self.current_call_id,
632
+ type='function',
633
+ function=DeltaFunctionCall(
634
+ name=None, arguments=json_continue))
635
+ ])
636
+ self._emit_delta(delta)
637
+ self.current_param_is_first = False
638
+
639
+ def _char_data(self, data: str):
640
+ """Handle XML character data events"""
641
+ if data and self.current_param_name:
642
+ # If preprocessing stage determines deferred parsing is needed,
643
+ # only cache character data, no streaming output
644
+ if self.defer_current_parameter:
645
+ original_data = data
646
+ if self.should_emit_end_newline:
647
+ original_data = '\n' + original_data
648
+ self.should_emit_end_newline = False
649
+ if original_data.endswith('\n'):
650
+ self.should_emit_end_newline = True
651
+ original_data = original_data[:-1]
652
+ self.current_param_value += original_data
653
+ return
654
+
655
+ param_type = self._get_param_type(self.current_param_name)
656
+
657
+ # Check if this is the first time receiving data for this parameter
658
+ # If this is the first packet of data and starts with \n, remove \n
659
+ if not self.current_param_value and data.startswith('\n'):
660
+ data = data[1:]
661
+
662
+ # Output start quote for string type (if not already output)
663
+ if (param_type
664
+ in ['string', 'str', 'text', 'varchar', 'char', 'enum']
665
+ and not self.start_quote_emitted):
666
+ quote_delta = DeltaMessage(tool_calls=[
667
+ DeltaToolCall(index=self.tool_call_index - 1,
668
+ id=self.current_call_id,
669
+ type='function',
670
+ function=DeltaFunctionCall(name=None,
671
+ arguments='"'))
672
+ ])
673
+ self._emit_delta(quote_delta)
674
+ self.start_quote_emitted = True
675
+
676
+ if not data:
677
+ return
678
+
679
+ original_data = data
680
+ # Delay output of trailing newline
681
+ if self.should_emit_end_newline:
682
+ original_data = '\n' + original_data
683
+ self.should_emit_end_newline = False
684
+ if original_data.endswith('\n'):
685
+ self.should_emit_end_newline = True
686
+ original_data = original_data[:-1]
687
+ self.current_param_value += original_data
688
+
689
+ # convert parameter value by param_type
690
+ converted_value = self._convert_param_value(
691
+ self.current_param_value, param_type)
692
+ output_data = self._convert_for_json_streaming(
693
+ converted_value, param_type)
694
+
695
+ delta_data = output_data[len(self.current_param_value_converted):]
696
+ self.current_param_value_converted = output_data
697
+
698
+ delta = DeltaMessage(tool_calls=[
699
+ DeltaToolCall(index=self.tool_call_index - 1,
700
+ id=self.current_call_id,
701
+ type='function',
702
+ function=DeltaFunctionCall(name=None,
703
+ arguments=delta_data))
704
+ ])
705
+ self._emit_delta(delta)
706
+
707
+ def _end_element(self, name: str):
708
+ """Handle XML end element events"""
709
+
710
+ if name == 'root':
711
+ return
712
+
713
+ # If function or tool_call ends and there are still unclosed parameters,
714
+ # complete parameter end first
715
+ if (name.startswith('function') or name == 'function'
716
+ or name == 'tool_call') and self.current_param_name:
717
+ self._auto_close_open_parameter_if_needed()
718
+
719
+ if (name.startswith('parameter')
720
+ or name == 'parameter') and self.current_param_name:
721
+ # End current parameter
722
+ param_name = self.current_param_name
723
+ param_value = self.current_param_value
724
+
725
+ # If in deferred parsing mode,
726
+ # perform overall parsing on raw content
727
+ # accumulated in preprocessing stage and output once
728
+ if self.defer_current_parameter:
729
+ raw_text = self.deferred_param_raw_value \
730
+ if self.deferred_param_raw_value else param_value
731
+ parsed_value = None
732
+ output_arguments = None
733
+ try:
734
+ # If previously delayed trailing newline,
735
+ # add it back before parsing
736
+ if self.should_emit_end_newline:
737
+ raw_for_parse = raw_text + '\n'
738
+ else:
739
+ raw_for_parse = raw_text
740
+ parsed_value = ast.literal_eval(raw_for_parse)
741
+ output_arguments = json.dumps(parsed_value,
742
+ ensure_ascii=False)
743
+ except Exception:
744
+ # Fallback: output as string as-is
745
+ output_arguments = json.dumps(raw_text, ensure_ascii=False)
746
+ parsed_value = raw_text
747
+
748
+ delta = DeltaMessage(tool_calls=[
749
+ DeltaToolCall(index=self.tool_call_index - 1,
750
+ id=self.current_call_id,
751
+ type='function',
752
+ function=DeltaFunctionCall(
753
+ name=None, arguments=output_arguments))
754
+ ])
755
+ self._emit_delta(delta)
756
+
757
+ # Clean up and store
758
+ self.should_emit_end_newline = False
759
+ self.parameters[param_name] = parsed_value
760
+ self.current_param_name = None
761
+ self.current_param_value = ""
762
+ self.current_param_value_converted = ""
763
+ self.start_quote_emitted = False
764
+ self.defer_current_parameter = False
765
+ self.deferred_param_raw_value = ""
766
+ return
767
+
768
+ param_type = self._get_param_type(param_name)
769
+
770
+ # convert complete parameter value by param_type
771
+ converted_value = self._convert_param_value(
772
+ param_value, param_type)
773
+
774
+ # Decide whether to add end quote based on parameter type
775
+ if param_type in [
776
+ 'string', 'str', 'text', 'varchar', 'char', 'enum'
777
+ ]:
778
+ # For empty string parameters, need special handling
779
+ if not param_value and not self.start_quote_emitted:
780
+ # No start quote output,
781
+ # directly output complete empty string
782
+ delta = DeltaMessage(tool_calls=[
783
+ DeltaToolCall(index=self.tool_call_index - 1,
784
+ id=self.current_call_id,
785
+ type='function',
786
+ function=DeltaFunctionCall(
787
+ name=None, arguments='""'))
788
+ ])
789
+ self._emit_delta(delta)
790
+ else:
791
+ # Non-empty parameter value, output end quote
792
+ delta = DeltaMessage(tool_calls=[
793
+ DeltaToolCall(index=self.tool_call_index - 1,
794
+ id=self.current_call_id,
795
+ type='function',
796
+ function=DeltaFunctionCall(
797
+ name=None, arguments='"'))
798
+ ])
799
+ self._emit_delta(delta)
800
+
801
+ self.should_emit_end_newline = False
802
+ # Store converted value
803
+ self.parameters[param_name] = converted_value
804
+ self.current_param_name = None
805
+ self.current_param_value = ''
806
+ self.current_param_value_converted = ''
807
+ self.start_quote_emitted = False
808
+
809
+ elif name.startswith('function') or name == 'function':
810
+ # if there are parameters, close JSON object
811
+ if self.parameters:
812
+ delta = DeltaMessage(tool_calls=[
813
+ DeltaToolCall(index=self.tool_call_index - 1,
814
+ id=self.current_call_id,
815
+ type='function',
816
+ function=DeltaFunctionCall(name=None,
817
+ arguments='}'))
818
+ ])
819
+ self._emit_delta(delta)
820
+ # return empty object
821
+ else:
822
+ delta = DeltaMessage(tool_calls=[
823
+ DeltaToolCall(index=self.tool_call_index - 1,
824
+ id=self.current_call_id,
825
+ type='function',
826
+ function=DeltaFunctionCall(name=None,
827
+ arguments='{}'))
828
+ ])
829
+ self._emit_delta(delta)
830
+ self.current_function_open = False
831
+
832
+ elif name == 'tool_call':
833
+ # Before ending tool_call,
834
+ # ensure function is closed to complete missing right brace
835
+ if self.current_function_open:
836
+ # If there are still unclosed parameters, close them first
837
+ if self.current_param_name:
838
+ self._end_element('parameter')
839
+ # Close function, ensure output '}' or '{}'
840
+ self._end_element('function')
841
+ # Final Delta
842
+ delta = DeltaMessage(tool_calls=[
843
+ DeltaToolCall(index=self.tool_call_index - 1,
844
+ id=self.current_call_id,
845
+ type='function',
846
+ function=DeltaFunctionCall(name=None,
847
+ arguments=''))
848
+ ])
849
+ self._emit_delta(delta)
850
+
851
+ # Check if there's text content to output (between tool_calls)
852
+ if self.text_content_buffer.strip():
853
+ text_delta = DeltaMessage(content=self.text_content_buffer)
854
+ self._emit_delta(text_delta)
855
+
856
+ self._reset_xml_parser_after_tool_call()
857
+
858
+ def setup_parser(self):
859
+ """Set up XML parser event handlers"""
860
+ self.parser.buffer_text = True
861
+ self.parser.StartElementHandler = self._start_element
862
+ self.parser.EndElementHandler = self._end_element
863
+ self.parser.CharacterDataHandler = self._char_data
864
+
865
+ def set_tools(self, tools: Union[list[ChatCompletionToolsParam], None]):
866
+ """Set tool configuration information"""
867
+ self.tools = tools
868
+
869
+ def _get_next_call_id(self):
870
+ """Generate unique call ID"""
871
+ return f'call_{uuid.uuid4().hex[:24]}'
872
+
873
+ def _extract_function_name(self, name: str,
874
+ attrs: dict[str, str]) -> Optional[str]:
875
+ """Extract function name from various formats"""
876
+ if attrs and 'name' in attrs:
877
+ return attrs['name']
878
+
879
+ if '=' in name:
880
+ parts = name.split('=', 1)
881
+ if len(parts) == 2 and parts[0] == 'function':
882
+ return parts[1]
883
+
884
+ return None
885
+
886
+ def _extract_parameter_name(self, name: str,
887
+ attrs: dict[str, str]) -> Optional[str]:
888
+ """Extract parameter name from various formats"""
889
+ if attrs and 'name' in attrs:
890
+ return attrs['name']
891
+
892
+ if '=' in name:
893
+ parts = name.split('=', 1)
894
+ if len(parts) == 2 and parts[0] == 'parameter':
895
+ return parts[1]
896
+
897
+ return None
898
+
899
+ def _get_param_type(self, param_name: str) -> str:
900
+ """Get parameter type based on tool configuration, defaults to string
901
+ Args:
902
+ param_name: Parameter name
903
+
904
+ Returns:
905
+ Parameter type
906
+ """
907
+ if not self.tools or not self.current_function_name:
908
+ return 'string'
909
+
910
+ for tool in self.tools:
911
+ if not hasattr(tool, 'type') or not (hasattr(
912
+ tool, 'function') and hasattr(tool.function, 'name')):
913
+ continue
914
+ if tool.type == 'function' and \
915
+ tool.function.name == self.current_function_name:
916
+ if not hasattr(tool.function, 'parameters'):
917
+ return 'string'
918
+ params = tool.function.parameters
919
+ if isinstance(params, dict) and 'properties' in params:
920
+ properties = params['properties']
921
+ if param_name in properties and isinstance(
922
+ properties[param_name], dict):
923
+ return self.repair_param_type(
924
+ str(properties[param_name].get('type', 'string')))
925
+ elif isinstance(params, dict) and param_name in params:
926
+ param_config = params[param_name]
927
+ if isinstance(param_config, dict):
928
+ return self.repair_param_type(
929
+ str(param_config.get('type', 'string')))
930
+ break
931
+ return 'string'
932
+
933
+ def repair_param_type(self, param_type: str) -> str:
934
+ """Repair unknown parameter types by treating them as string
935
+ Args:
936
+ param_type: Parameter type
937
+
938
+ Returns:
939
+ Repaired parameter type
940
+ """
941
+ if param_type in [
942
+ 'string', 'str', 'text', 'varchar', 'char', 'enum'
943
+ ] or param_type.startswith('int') or param_type.startswith(
944
+ 'uint'
945
+ ) or param_type.startswith('long') or param_type.startswith(
946
+ 'short'
947
+ ) or param_type.startswith('unsigned') or param_type.startswith(
948
+ 'num') or param_type.startswith('float') or param_type in [
949
+ 'boolean', 'bool', 'binary'
950
+ ] or (param_type in ["object", "array", "arr", "sequence"]
951
+ or param_type.startswith("dict")
952
+ or param_type.startswith("list")):
953
+ return param_type
954
+ else:
955
+ return 'string'
956
+
957
+ def _convert_param_value(self, param_value: str, param_type: str) -> Any:
958
+ """Convert value based on parameter type
959
+ Args:
960
+ param_value: Parameter value
961
+ param_type: Parameter type
962
+
963
+ Returns:
964
+ Converted value
965
+ """
966
+ if param_value.lower() == 'null':
967
+ return None
968
+
969
+ param_type = param_type.strip().lower()
970
+ if param_type in ['string', 'str', 'text', 'varchar', 'char', 'enum']:
971
+ return param_value
972
+ elif (param_type.startswith('int') or param_type.startswith('uint')
973
+ or param_type.startswith('long')
974
+ or param_type.startswith('short')
975
+ or param_type.startswith('unsigned')):
976
+ try:
977
+ return int(param_value)
978
+ except (ValueError, TypeError):
979
+ logger.warning(
980
+ "Parsed value '%s' of parameter '%s' is not an integer "
981
+ "in tool '%s', degenerating to string.", param_value)
982
+ return param_value
983
+ elif param_type.startswith('num') or param_type.startswith('float'):
984
+ try:
985
+ float_param_value: float = float(param_value)
986
+ return float_param_value if float_param_value - int(
987
+ float_param_value) != 0 else int(float_param_value)
988
+ except (ValueError, TypeError):
989
+ logger.warning(
990
+ "Parsed value '%s' of parameter '%s' is not a float "
991
+ "in tool '%s', degenerating to string.", param_value)
992
+ return param_value
993
+ elif param_type in ['boolean', 'bool', 'binary']:
994
+ param_value = param_value.lower()
995
+ return param_value == 'true'
996
+ else:
997
+ return param_value
998
+
999
+ def _convert_for_json_streaming(self, converted_value: Any,
1000
+ param_type: str) -> str:
1001
+ """Convert converted_value based on
1002
+ whether it's empty and if type is string
1003
+ Args:
1004
+ converted_value: Converted value
1005
+ param_type: Parameter type
1006
+
1007
+ Returns:
1008
+ Converted string for streaming output
1009
+ """
1010
+ # Check if value is empty, but exclude numeric 0
1011
+ if converted_value is None or converted_value == '':
1012
+ return ''
1013
+
1014
+ if param_type in ['string', 'str', 'text', 'varchar', 'char', 'enum']:
1015
+ # String type, remove double quotes
1016
+ return json.dumps(converted_value, ensure_ascii=False)[1:-1]
1017
+ else:
1018
+ # Non-string type, return complete JSON string
1019
+ if not isinstance(converted_value, str):
1020
+ return json.dumps(converted_value, ensure_ascii=False)
1021
+ else:
1022
+ return converted_value
1023
+
1024
+ def _reset_xml_parser_after_tool_call(self):
1025
+ """
1026
+ Each tool_call is treated as a separate XML document,
1027
+ so we need to reset the parser after each tool_call.
1028
+ """
1029
+
1030
+ # recreate XML parser
1031
+ self.parser = ParserCreate()
1032
+ self.setup_parser()
1033
+
1034
+ # Reset current tool_call state
1035
+ if self.current_call_id:
1036
+ self.last_completed_call_id = self.current_call_id
1037
+ self.current_call_id = None
1038
+ self.current_function_name = None
1039
+ self.current_function_open = False
1040
+ self.parameters = {}
1041
+ self.current_param_name = None
1042
+ self.current_param_value = ''
1043
+ self.current_param_value_converted = ''
1044
+ self.current_param_is_first = False
1045
+ self.should_emit_end_newline = False
1046
+ self.start_quote_emitted = False
1047
+ self.text_content_buffer = ''
1048
+
1049
+ # Reset preprocessing and deferred parsing state
1050
+ self._pre_inside_parameter = False
1051
+ self._pre_param_buffer = ""
1052
+ self._pre_current_param_name = None
1053
+ self.defer_current_parameter = False
1054
+ self.deferred_param_raw_value = ""
1055
+
1056
+
1057
+ @ToolParserManager.register_module("qwen3_xml")
1058
+ class Qwen3XMLToolParser(ToolParser):
1059
+
1060
+ def __init__(self, tokenizer: AnyTokenizer):
1061
+ super().__init__(tokenizer)
1062
+ self.parser = StreamingXMLToolCallParser()
1063
+
1064
+ logger.info("vLLM Successfully import tool parser %s !",
1065
+ self.__class__.__name__)
1066
+
1067
+ def extract_tool_calls(
1068
+ self,
1069
+ model_output: str,
1070
+ request: ChatCompletionRequest,
1071
+ ) -> ExtractedToolCallInformation:
1072
+ self.parser.reset_streaming_state()
1073
+ if request:
1074
+ self.parser.set_tools(request.tools)
1075
+ result = self.parser.parse_single_streaming_chunks(model_output)
1076
+ if not result.tool_calls:
1077
+ return ExtractedToolCallInformation(
1078
+ tool_calls=[],
1079
+ tools_called=False,
1080
+ content=result.content,
1081
+ )
1082
+ else:
1083
+ tool_calls = []
1084
+ for tool_call in result.tool_calls:
1085
+ if tool_call.function and tool_call.function.name:
1086
+ tool_calls.append(
1087
+ ToolCall(
1088
+ id=tool_call.id,
1089
+ type=tool_call.type,
1090
+ function=FunctionCall(
1091
+ name=tool_call.function.name,
1092
+ arguments=tool_call.function.arguments,
1093
+ ),
1094
+ ))
1095
+ return ExtractedToolCallInformation(
1096
+ tool_calls=tool_calls,
1097
+ tools_called=len(tool_calls) > 0,
1098
+ content=result.content,
1099
+ )
1100
+
1101
+ def extract_tool_calls_streaming(
1102
+ self,
1103
+ previous_text: str,
1104
+ current_text: str,
1105
+ delta_text: str,
1106
+ previous_token_ids: Sequence[int],
1107
+ current_token_ids: Sequence[int],
1108
+ delta_token_ids: Sequence[int],
1109
+ request: ChatCompletionRequest,
1110
+ ) -> Union[DeltaMessage, None]:
1111
+ if not previous_text:
1112
+ self.parser.reset_streaming_state()
1113
+ if request:
1114
+ self.parser.set_tools(request.tools)
1115
+
1116
+ # Model sometimes outputs separately causing delta_text to be empty.
1117
+ # If there were tool_calls before and all current tool_calls have ended,
1118
+ # return an empty tool_call for outer streaming output
1119
+ # to correctly output tool_call field
1120
+ if not delta_text and delta_token_ids:
1121
+ open_calls = current_text.count(
1122
+ self.parser.tool_call_start_token) - current_text.count(
1123
+ self.parser.tool_call_end_token)
1124
+ if open_calls == 0 and self.parser.tool_call_index > 0:
1125
+ # If current_call_id is None, use last_completed_call_id
1126
+ call_id = self.parser.current_call_id or \
1127
+ self.parser.last_completed_call_id
1128
+ return DeltaMessage(tool_calls=[
1129
+ DeltaToolCall(
1130
+ index=self.parser.tool_call_index - 1,
1131
+ id=call_id,
1132
+ function=DeltaFunctionCall(arguments=''),
1133
+ type='function',
1134
+ )
1135
+ ])
1136
+
1137
+ return self.parser.parse_single_streaming_chunks(delta_text)