For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Python class
MultiGraphPipelineModelWithKVCache
MultiGraphPipelineModelWithKVCache
class max.pipelines.lib.MultiGraphPipelineModelWithKVCache(pipeline_config, session, devices, kv_cache_config, weights, adapter, return_logits, *, memory_plan, return_hidden_states=ReturnHiddenStates.NONE, max_batch_size=1)
Bases: PipelineModelWithKVCache[BaseContextType, _ModelConfigT, Model | None]
Graph-API VLM with unified load_model() and per-tower hooks.
_create_model_config() should return the full VLM config (with
.vision_config and .text_config / .llm_config subconfigs) and
assign model_config. Both _build_*() hooks receive that same
model_config.
Override load_model() when graph capture or weight loading does not
fit this flow (such as Qwen2.5VL, Kimi-K2.5).
-
Parameters:
-
- pipeline_config (PipelineConfig)
- session (InferenceSession)
- devices (list[Device])
- kv_cache_config (KVCacheConfig)
- weights (Weights)
- adapter (WeightsAdapter | None)
- return_logits (ReturnLogits)
- memory_plan (MemoryPlan)
- return_hidden_states (ReturnHiddenStates)
- max_batch_size (int)
load_model()
load_model(session)
Build, compile, and load vision and language graphs into session.
-
Parameters:
-
session (InferenceSession)
-
Return type: