Skip to main content

/

DocsCode

Python module

max.pipelines.architectures.qwen2_5vl

Qwen2.5-VL vision-language architecture for multimodal text generation.

Qwen2_5VLConfig​

class max.pipelines.architectures.qwen2_5vl.Qwen2_5VLConfig(*, devices, image_token_id, video_token_id, vision_start_token_id, spatial_merge_size, tokens_per_second, mrope_section, vision_config, llm_config)

source

Bases: ArchConfigWithKVCache

Configuration for Qwen2.5VL models.

Parameters:

calculate_max_seq_len()​

static calculate_max_seq_len(pipeline_config, huggingface_config)

source

Calculate maximum sequence length for Qwen2.5VL.

Parameters:

Return type:

int

construct_kv_params()​

static construct_kv_params(huggingface_config, pipeline_config, devices, kv_cache_config, cache_dtype)

source

Parameters:

Return type:

KVCacheParams

devices​

devices: list[DeviceRef]

source

Devices that the Qwen2.5VL model is parallelized over.

finalize()​

finalize(huggingface_config, pipeline_config, llm_state_dict, vision_state_dict, return_logits, norm_method='rms_norm')

source

Finalize the Qwen2_5VLConfig instance with state_dict dependent fields.

Parameters:

  • huggingface_config (AutoConfig) – HuggingFace model configuration.
  • pipeline_config (PipelineConfig) – The MAX Engine pipeline configuration.
  • llm_state_dict (dict[str, WeightData]) – Language model weights dictionary.
  • vision_state_dict (dict[str, WeightData]) – Vision encoder weights dictionary.
  • return_logits (ReturnLogits) – Return logits configuration.
  • norm_method (Literal['rms_norm', 'layer_norm']) – Normalization method.

Return type:

None

get_kv_params()​

get_kv_params()

source

Returns the KV cache parameters from the embedded LLM config.

Return type:

KVCacheParams

get_max_seq_len()​

get_max_seq_len()

source

Returns the maximum sequence length from the embedded LLM config.

Return type:

int

get_num_layers()​

static get_num_layers(huggingface_config)

source

Parameters:

huggingface_config (AutoConfig)

Return type:

int

image_token_id​

image_token_id: int

source

Token ID used for image placeholders in the input sequence.

initialize()​

classmethod initialize(pipeline_config, model_config=None)

source

Initializes a Qwen2_5VLConfig instance from pipeline configuration.

Parameters:

Returns:

A Qwen2_5VLConfig instance with fields initialized from config.

Return type:

Self

initialize_from_config()​

classmethod initialize_from_config(pipeline_config, huggingface_config)

source

Initializes a Qwen2_5VLConfig from pipeline and HuggingFace configs.

This method creates a config instance with all fields that can be determined from the pipeline and HuggingFace configurations, without needing the state_dict. Fields that depend on the state_dict should be set via the finalize() method.

Parameters:

  • pipeline_config (PipelineConfig) – The MAX Engine pipeline configuration.
  • huggingface_config (AutoConfig) – HuggingFace model configuration.

Returns:

A Qwen2_5VLConfig instance ready for finalization.

Return type:

Self

llm_config​

llm_config: Llama3Config

source

Language model configuration using Llama3 architecture.

mrope_section​

mrope_section: list[int]

source

List of indices for the mrope section.

spatial_merge_size​

spatial_merge_size: int

source

Size parameter for spatial merging of vision features.

tokens_per_second​

tokens_per_second: int

source

Number of tokens per second.

video_token_id​

video_token_id: int

source

Token ID used for video placeholders in the input sequence.

vision_config​

vision_config: VisionConfig

source

Vision encoder configuration.

vision_start_token_id​

vision_start_token_id: int

source

Token ID that marks the start of vision content.

Qwen2_5VLModel​

class max.pipelines.architectures.qwen2_5vl.Qwen2_5VLModel(pipeline_config, session, devices, kv_cache_config, weights, adapter=None, return_logits=ReturnLogits.LAST_TOKEN)

source

Bases: AlwaysSignalBuffersMixin, PipelineModelWithKVCache[TextAndVisionContext]

A Qwen2.5VL pipeline model for multimodal text generation.

Parameters:

calculate_max_seq_len()​

static calculate_max_seq_len(pipeline_config, huggingface_config)

source

Calculates the maximum sequence length for the Qwen2.5VL model.

Parameters:

Return type:

int

estimate_activation_memory()​

classmethod estimate_activation_memory(pipeline_config, huggingface_config)

source

Estimates the activation memory required for model execution.

This accounts for temporary memory buffers used during model execution, such as intermediate activations and working buffers.

The default implementation returns 0 for backward compatibility. Models with significant activation memory requirements should override this method to provide accurate estimates.

Parameters:

  • pipeline_config (PipelineConfig) – Pipeline configuration
  • huggingface_config (AutoConfig) – Hugging Face model configuration

Returns:

Estimated activation memory in bytes

Return type:

int

execute()​

execute(model_inputs)

source

Executes the Qwen2.5VL model with the prepared inputs.

Parameters:

model_inputs (ModelInputs)

Return type:

ModelOutputs

get_kv_params()​

classmethod get_kv_params(huggingface_config, pipeline_config, devices, kv_cache_config, cache_dtype)

source

Gets the parameters required to configure the KV cache for Qwen2.5VL.

Parameters:

Return type:

KVCacheParams

language_model​

language_model: Model

source

The compiled language model for text generation.

load_model()​

load_model(session)

source

Loads the compiled Qwen2.5VL models into the MAX Engine session.

Returns:

A tuple of (vision_model, language_model).

Parameters:

session (InferenceSession)

Return type:

tuple[Model, Model]

model_config​

model_config: Qwen2_5VLConfig | None

source

The Qwen2.5VL model configuration.

prepare_decoder_position_ids()​

static prepare_decoder_position_ids(context_batch, devices)

source

Prepare decoder position IDs for a batch of contexts.

This function computes position IDs for decoder tokens, handling three cases:

  1. Vision encoding with pre-computed position IDs (use stored values)
  2. Vision encoding requiring recomputation (after preemption)
  3. Text-only generation (simple arange with offset)

Optimized implementation: pre-allocates output array and writes directly, avoiding concatenation overhead for better performance.

Parameters:

  • context_batch (Sequence[Qwen2_5VLTextAndVisionContext]) – Sequence of Qwen2.5VL contexts to process
  • devices (list[Device]) – List of devices to place the output tensor on

Returns:

Buffer containing decoder position IDs with shape [n_rope_sections, total_seq_len]

Return type:

Buffer

prepare_initial_token_inputs()​

prepare_initial_token_inputs(replica_batches, kv_cache_inputs=None, return_n_logits=1)

source

Prepares the initial inputs for the first execution pass of the Qwen2.5VL model.

Parameters:

Return type:

Qwen2_5VLInputs

prepare_next_token_inputs()​

prepare_next_token_inputs(next_tokens, prev_model_inputs)

source

Prepares the inputs for subsequent execution steps in a multi-step generation.

Parameters:

Return type:

Qwen2_5VLInputs

vision_model​

vision_model: Model

source

The compiled vision model for processing images.

Models

Cookbook

Puzzles

Blog

Community

Contact