o :/i @sJddlmZmZddlmZddlmZddlmZeGdddZ dS))AnyLiteral)field_validator)config)AttentionBackendEnumc@seZdZUdZdZedBed< dZeddBed< dZ e ed< dZ e ed < dZ e ed < dZe ed < dZe dBed < d Ze ed< dZe ed< dZe ed< defddZedddededefddZdS)AttentionConfigz/Configuration for attention mechanisms in vLLM.Nbackend)flash_attn_versionFuse_prefill_decode_attention (flash_attn_max_num_splits_for_cuda_graphuse_cudnn_prefill"use_trtllm_ragged_deepseek_prefilluse_trtllm_attentionTdisable_flashinfer_prefill!disable_flashinfer_q_quantizationuse_prefill_query_quantizationreturncCs&ddlm}m}g}|||}||S)a$ Provide a hash that uniquely identifies all the configs that affect the structure of the computation graph from input ids/embeddings to the final hidden states, excluding anything before input ids/embeddings and after the final hidden states. r)get_hash_factors hash_factors)vllm.config.utilsrr)selfrrignored_factorsfactorsrb/lsinfo/ai/hellotax_ai/llm_service/venv_vllm/lib/python3.10/site-packages/vllm/config/attention.py compute_hash1s zAttentionConfig.compute_hashbefore)modevaluecCs*t|tr|dkr dSt|S|S)zEnable parsing of the `backend` enum type from string. The special value "auto" is treated as None, which triggers automatic backend selection. autoN) isinstancestrlowerrupper)clsr"rrrvalidate_backend_before?s   z'AttentionConfig.validate_backend_before)__name__ __module__ __qualname____doc__rr__annotations__r rr boolrintrrrrrrr%rr classmethodrr)rrrrr s4         rN) typingrrpydanticrrr#vllm.v1.attention.backends.registryrrrrrrs