o ٷi@sdZddlmZddlmZddlmZddlmZddl m Z m Z ddl m Z eeZded e d d fd d ZdeiZGdddeZd S)z TeaCache backend implementation. This module provides the TeaCache backend that implements the CacheBackend interface using the hooks-based TeaCache system. )Any) init_logger) CacheBackend)TeaCacheConfig) TeaCacheHookapply_teacache_hook)DiffusionCacheConfigpipelineconfigreturnNcsvtd|j|jd}|j}|jddl}fdd}||||_t|||j|_||_ t d|jd|j dS) z* Enable TeaCache for Bagel model. Bageltransformer_type rel_l1_thresh coefficientsrNcs|i|S)N)selfargskwargsoriginal_forward_flowr^/home/ubuntu/.local/lib/python3.10/site-packages/vllm_omni/diffusion/cache/teacache/backend.py forward_alias%sz,enable_bagel_teacache..forward_alias$TeaCache applied with rel_l1_thresh=, transformer_class=) rrrbagel _forward_flowtypes MethodTypeforwardr transformerloggerinfor)r r teacache_configr rrrrrenable_bagel_teacaches$   r$ BagelPipelinec @s>eZdZdZdeddfddZd deded eddfd d ZdS) TeaCacheBackenda+ TeaCache implementation using hooks. TeaCache (Timestep Embedding Aware Cache) is an adaptive caching technique that speeds up diffusion inference by reusing transformer block computations when consecutive timestep embeddings are similar. The backend applies TeaCache hooks to the transformer which intercept the forward pass and implement the caching logic transparently. Example: >>> from vllm_omni.diffusion.data import DiffusionCacheConfig >>> backend = TeaCacheBackend(DiffusionCacheConfig(rel_l1_thresh=0.2)) >>> backend.enable(pipeline) >>> # Generate with cache enabled >>> backend.refresh(pipeline, num_inference_steps=50) # Refresh before each generation >>> # Access config attributes: backend.config.rel_l1_thresh r r Nc Cs|jj}|tvrtd|t|||jnC|j}|jj}z t||jj|jj d}Wnt yI}zt d|t d|dd}~wwt ||td|jd|jd |_dS) a Enable TeaCache on transformer using hooks. This creates a TeaCacheConfig from the backend's DiffusionCacheConfig and applies the TeaCache hook to the transformer. Args: pipeline: Diffusion pipeline instance. Extracts transformer and transformer_type: - transformer: pipeline.transformer - transformer_type: pipeline.transformer.__class__.__name__ z)Using custom TeaCache enabler for model: r z!Failed to create TeaCacheConfig: z Invalid TeaCache configuration: z. Expected keys: rel_l1_thresh, coefficients (optional). transformer_type is automatically extracted from pipeline.transformer.__class__.__name__.NrrT) __class____name__CUSTOM_TEACACHE_ENABLERSr!r"r r rrr Exceptionerror ValueErrorrrenabled)rr pipeline_typer rr#errrenableJs4      zTeaCacheBackend.enableTnum_inference_stepsverbosecCs~|j}t|dr4|jtj}|dur)|jtj|r'td|ddSdS|r2t ddSdS|r=t ddSdS)a@ Refresh TeaCache state for new generation. Clears all cached residuals and resets counters/accumulators. Should be called before each generation to ensure clean state. Args: pipeline: Diffusion pipeline instance. Extracts transformer via pipeline.transformer. num_inference_steps: Number of inference steps for the current generation. Currently not used by TeaCache but accepted for interface consistency. verbose: Whether to log refresh operations (default: True) _hook_registryNz.TeaCache state refreshed (num_inference_steps=)z+TeaCache hook not found, nothing to refreshz=Transformer has no hook registry, TeaCache may not be applied) r hasattrr3get_hookr _HOOK_NAME reset_hookr!debugwarning)rr r1r2r hookrrrrefresh}s zTeaCacheBackend.refresh)T) r( __module__ __qualname____doc__rr0intboolr<rrrrr&6s 3r&)r?typingr vllm.loggerrvllm_omni.diffusion.cache.baser)vllm_omni.diffusion.cache.teacache.configr'vllm_omni.diffusion.cache.teacache.hookrrvllm_omni.diffusion.datarr(r!r$r)r&rrrrs