# MAX Python API Documentation

> The MAX Python API reference (package indexes).

Version: Nightly

This file contains links to documentation sections following the llmstxt.org standard.

## Table of Contents

- [max.driver](https://docs.modular.com/api/python/driver/): | [`Accelerator`](generated/max.driver.Accelerator.md#max.driver.Accelerator)          | Creates an accelerator device with the specified ID and me...
- [max.dtype](https://docs.modular.com/api/python/dtype/): | [`DType`](generated/max.dtype.DType.md#max.dtype.DType) | The tensor data type.                                                                  ...
- [max.engine](https://docs.modular.com/api/python/engine/): Modular engine provides methods to load and execute AI models.
- [max.experimental.functional](https://docs.modular.com/api/python/experimental.functional/): Distributed functional ops with explicit per-op SPMD dispatch.
- [max.experimental](https://docs.modular.com/api/python/experimental/): Experimental APIs for building, sharding, and running ML workloads.
- [max.experimental.nn](https://docs.modular.com/api/python/experimental.nn/): Module framework for [`max.experimental`](experimental.md#module-max.experimental).
- [max.experimental.nn.norm](https://docs.modular.com/api/python/experimental.nn.norm/): Normalization layers for MAX neural networks.
- [max.experimental.nn.rope](https://docs.modular.com/api/python/experimental.nn.rope/): Positional embedding modules and functions.
- [max.experimental.sharding](https://docs.modular.com/api/python/experimental.sharding/): Distributed-tensor sharding: how a tensor is laid out across a device mesh.
- [max.experimental.tensor](https://docs.modular.com/api/python/experimental.tensor/): Provides tensor operations with eager execution capabilities.
- [max.experimental.testing](https://docs.modular.com/api/python/experimental.testing/): Testing utilities for the experimental [`Tensor`](generated/max.experimental.tensor.Tensor.md#max.experimental.tensor.Tensor) API.
- [max.experimental.torch](https://docs.modular.com/api/python/experimental.torch/): | [`CustomOpLibrary`](generated/max.experimental.torch.CustomOpLibrary.md#max.experimental.torch.CustomOpLibrary) | A PyTorch interface to custom o...
- [max.graph](https://docs.modular.com/api/python/graph/): APIs to build inference graphs for MAX.
- [max.graph.ops](https://docs.modular.com/api/python/graph.ops/): Implements operations used when staging a graph.
- [max.graph.quantization](https://docs.modular.com/api/python/graph.quantization/): Defines quantization encodings and configuration types for MAX graph weights.
- [max.graph.weights](https://docs.modular.com/api/python/graph.weights/): | [`WeightData`](generated/max.graph.weights.WeightData.md#max.graph.weights.WeightData)             | Container for weight tensor data with metada...
- [max](https://docs.modular.com/api/python/): The MAX Python API reference.
- [max.nn.attention](https://docs.modular.com/api/python/nn.attention/): The attention mechanism used within the model.
- [max.nn.kernels](https://docs.modular.com/api/python/nn.kernels/): Helper functions for wrapping custom kv cache/attention related ops.
- [max.nn.kv_cache](https://docs.modular.com/api/python/nn.kv_cache/): | [`KVCacheBuffer`](generated/max.nn.kv_cache.KVCacheBuffer.md#max.nn.kv_cache.KVCacheBuffer)                                     | A collection of...
- [max.nn](https://docs.modular.com/api/python/nn/): Neural network modules for MAX.
- [max.pipelines.architectures.bert](https://docs.modular.com/api/python/pipelines.architectures.bert/): BERT sentence transformer architecture for embeddings generation.
- [max.pipelines.architectures.deepseekV2](https://docs.modular.com/api/python/pipelines.architectures.deepseekV2/): DeepSeek-V2 mixture-of-experts architecture for text generation.
- [max.pipelines.architectures.deepseekV3](https://docs.modular.com/api/python/pipelines.architectures.deepseekV3/): DeepSeek-V3 mixture-of-experts architecture for text generation.
- [max.pipelines.architectures.deepseekV3_2](https://docs.modular.com/api/python/pipelines.architectures.deepseekV3_2/): DeepSeek-V3.2 mixture-of-experts architecture for text generation.
- [max.pipelines.architectures.deepseekV3_nextn](https://docs.modular.com/api/python/pipelines.architectures.deepseekV3_nextn/): DeepSeek-V3 NextN multi-token prediction draft model for speculative decoding.
- [max.pipelines.architectures.dflash_llama3](https://docs.modular.com/api/python/pipelines.architectures.dflash_llama3/): DFlash draft model for Llama3-family targets.
- [max.pipelines.architectures.diffusion_gemma](https://docs.modular.com/api/python/pipelines.architectures.diffusion_gemma/): DiffusionGemma block-diffusion architecture.
- [max.pipelines.architectures.eagle3_deepseekV3](https://docs.modular.com/api/python/pipelines.architectures.eagle3_deepseekV3/): DeepseekV3 + Eagle3 speculator pipeline.
- [max.pipelines.architectures.eagle_llama3](https://docs.modular.com/api/python/pipelines.architectures.eagle_llama3/): EAGLE speculative decoding draft model for Llama 3.
- [max.pipelines.architectures.flux2](https://docs.modular.com/api/python/pipelines.architectures.flux2/): FLUX.2 diffusion architecture for image generation.
- [max.pipelines.architectures.gemma3](https://docs.modular.com/api/python/pipelines.architectures.gemma3/): Gemma 3 transformer architecture for text generation.
- [max.pipelines.architectures.gemma3multimodal](https://docs.modular.com/api/python/pipelines.architectures.gemma3multimodal/): Gemma 3 vision-language architecture for multimodal text generation.
- [max.pipelines.architectures.gemma4](https://docs.modular.com/api/python/pipelines.architectures.gemma4/): Gemma 4 vision-language architecture for multimodal text generation.
- [max.pipelines.architectures.gemma4_assistant](https://docs.modular.com/api/python/pipelines.architectures.gemma4_assistant/): <blockquote>
- [max.pipelines.architectures.glm5_1](https://docs.modular.com/api/python/pipelines.architectures.glm5_1/): GLM-5.1 (GlmMoeDsa) mixture-of-experts architecture for text generation.
- [max.pipelines.architectures.gpt_oss](https://docs.modular.com/api/python/pipelines.architectures.gpt_oss/): GPT-OSS mixture-of-experts architecture for text generation.
- [max.pipelines.architectures.granite](https://docs.modular.com/api/python/pipelines.architectures.granite/): Granite architecture, builds on [`llama3`](pipelines.architectures.llama3.md#module-max.pipelines.architectures.llama3).
- [max.pipelines.architectures.hy_v3](https://docs.modular.com/api/python/pipelines.architectures.hy_v3/): Tencent Hunyuan Hy3-preview (HYV3ForCausalLM).
- [max.pipelines.architectures.idefics3](https://docs.modular.com/api/python/pipelines.architectures.idefics3/): Idefics3 vision-language architecture for multimodal text generation.
- [max.pipelines.architectures.ideogram4](https://docs.modular.com/api/python/pipelines.architectures.ideogram4/): Ideogram 4 flow-matching text-to-image architecture.
- [max.pipelines.architectures.internvl](https://docs.modular.com/api/python/pipelines.architectures.internvl/): InternVL vision-language architecture for multimodal text generation.
- [max.pipelines.architectures.kimik2_5](https://docs.modular.com/api/python/pipelines.architectures.kimik2_5/): Kimi K2.5 mixture-of-experts architecture for text generation.
- [max.pipelines.architectures.laguna](https://docs.modular.com/api/python/pipelines.architectures.laguna/): <blockquote>
- [max.pipelines.architectures.lfm2](https://docs.modular.com/api/python/pipelines.architectures.lfm2/): <blockquote>
- [max.pipelines.architectures.llama3](https://docs.modular.com/api/python/pipelines.architectures.llama3/): Llama 3 transformer architecture for text generation.
- [max.pipelines.architectures.llama4](https://docs.modular.com/api/python/pipelines.architectures.llama4/): Llama 4 (text-only) transformer architecture for text generation.
- [max.pipelines.architectures.mamba](https://docs.modular.com/api/python/pipelines.architectures.mamba/): Mamba state-space architecture for text generation.
- [max.pipelines.architectures](https://docs.modular.com/api/python/pipelines.architectures/): MAX includes built-in support for a wide range of model architectures. Each
- [max.pipelines.architectures.minimax_m2](https://docs.modular.com/api/python/pipelines.architectures.minimax_m2/): <blockquote>
- [max.pipelines.architectures.mistral](https://docs.modular.com/api/python/pipelines.architectures.mistral/): Mistral transformer architecture for text generation.
- [max.pipelines.architectures.mistral3](https://docs.modular.com/api/python/pipelines.architectures.mistral3/): Mistral 3 vision-language architecture for multimodal text generation.
- [max.pipelines.architectures.mpnet](https://docs.modular.com/api/python/pipelines.architectures.mpnet/): MPNet sentence transformer architecture for embeddings generation.
- [max.pipelines.architectures.nemotron_h](https://docs.modular.com/api/python/pipelines.architectures.nemotron_h/): <blockquote>
- [max.pipelines.architectures.olmo](https://docs.modular.com/api/python/pipelines.architectures.olmo/): OLMo transformer architecture for text generation.
- [max.pipelines.architectures.olmo2](https://docs.modular.com/api/python/pipelines.architectures.olmo2/): OLMo 2 transformer architecture for text generation.
- [max.pipelines.architectures.olmo3](https://docs.modular.com/api/python/pipelines.architectures.olmo3/): OLMo 3 transformer architecture for text generation.
- [max.pipelines.architectures.phi3](https://docs.modular.com/api/python/pipelines.architectures.phi3/): Phi-3 transformer architecture for text generation.
- [max.pipelines.architectures.pixtral](https://docs.modular.com/api/python/pipelines.architectures.pixtral/): Pixtral vision-language architecture for multimodal text generation.
- [max.pipelines.architectures.qwen2](https://docs.modular.com/api/python/pipelines.architectures.qwen2/): Qwen2 transformer architecture for text generation.
- [max.pipelines.architectures.qwen2_5vl](https://docs.modular.com/api/python/pipelines.architectures.qwen2_5vl/): Qwen2.5-VL vision-language architecture for multimodal text generation.
- [max.pipelines.architectures.qwen3](https://docs.modular.com/api/python/pipelines.architectures.qwen3/): Qwen3 transformer architecture for text generation.
- [max.pipelines.architectures.qwen3_5](https://docs.modular.com/api/python/pipelines.architectures.qwen3_5/): <blockquote>
- [max.pipelines.architectures.qwen3_embedding](https://docs.modular.com/api/python/pipelines.architectures.qwen3_embedding/): Qwen3 architecture for embeddings generation.
- [max.pipelines.architectures.qwen3vl_moe](https://docs.modular.com/api/python/pipelines.architectures.qwen3vl_moe/): Qwen3-VL vision-language architecture for multimodal text generation.
- [max.pipelines.architectures.qwen_image](https://docs.modular.com/api/python/pipelines.architectures.qwen_image/): Qwen-Image diffusion architecture for image generation.
- [max.pipelines.architectures.qwen_image_edit](https://docs.modular.com/api/python/pipelines.architectures.qwen_image_edit/): Qwen-Image-Edit diffusion architecture for image editing.
- [max.pipelines.architectures.step3p5](https://docs.modular.com/api/python/pipelines.architectures.step3p5/): <blockquote>
- [max.pipelines.architectures.unified_dflash_kimi_k25](https://docs.modular.com/api/python/pipelines.architectures.unified_dflash_kimi_k25/): DFlash speculative decoding for Kimi K2.5 with unified graph compilation.
- [max.pipelines.architectures.unified_dflash_llama3](https://docs.modular.com/api/python/pipelines.architectures.unified_dflash_llama3/): DFlash speculative decoding for Llama3 with unified graph compilation.
- [max.pipelines.architectures.unified_eagle_llama3](https://docs.modular.com/api/python/pipelines.architectures.unified_eagle_llama3/): EAGLE speculative decoding draft model for Llama 3 with unified graph compilation.
- [max.pipelines.architectures.unified_mtp_deepseekV3](https://docs.modular.com/api/python/pipelines.architectures.unified_mtp_deepseekV3/): DeepSeek-V3 multi-token prediction draft model for speculative decoding with unified graph compilation.
- [max.pipelines.architectures.unified_mtp_gemma4](https://docs.modular.com/api/python/pipelines.architectures.unified_mtp_gemma4/): Gemma4 with MTP draft model for speculative decoding with unified graph compilation.
- [max.pipelines.architectures.unified_mtp_glm5_2](https://docs.modular.com/api/python/pipelines.architectures.unified_mtp_glm5_2/): GLM-5.2 (DeepSeek-V3.2 sparse) MTP draft model for unified speculative decoding.
- [max.pipelines.architectures.wan](https://docs.modular.com/api/python/pipelines.architectures.wan/): Wan diffusion architecture for video generation.
- [max.pipelines.context](https://docs.modular.com/api/python/pipelines.context/): | [`TextContext`](generated/max.pipelines.context.TextContext.md#max.pipelines.context.TextContext)                            | A base class for m...
- [max.pipelines.diffusion](https://docs.modular.com/api/python/pipelines.diffusion/): Diffusion-specific pipeline components.
- [max.pipelines.diffusion.schedulers](https://docs.modular.com/api/python/pipelines.diffusion.schedulers/): | [`FlowMatchEulerDiscreteScheduler`](generated/max.pipelines.diffusion.schedulers.FlowMatchEulerDiscreteScheduler.md#max.pipelines.diffusion.sched...
- [max.pipelines.kv_cache](https://docs.modular.com/api/python/pipelines.kv_cache/): KV cache management for MAX pipelines.
- [max.pipelines.lib.interfaces](https://docs.modular.com/api/python/pipelines.lib.interfaces/): Interfaces for MAX pipelines.
- [max.pipelines.lib.log_probabilities](https://docs.modular.com/api/python/pipelines.lib.log_probabilities/): Builds computation graphs for log probabilities over batched input sequences.
- [max.pipelines.lib](https://docs.modular.com/api/python/pipelines.lib/): Types to interface with ML pipelines such as text/token generation.
- [max.pipelines.lib.registry](https://docs.modular.com/api/python/pipelines.lib.registry/): Model registry, for tracking various model variants.
- [max.pipelines.logging_utils](https://docs.modular.com/api/python/pipelines.logging_utils/): Pipeline logging utilities.
- [max.pipelines.lora](https://docs.modular.com/api/python/pipelines.lora/): LoRA adapter management for MAX pipelines.
- [max.pipelines](https://docs.modular.com/api/python/pipelines/): Types to interface with ML pipelines such as text/token/pixel generation.
- [max.pipelines.modeling.base](https://docs.modular.com/api/python/pipelines.modeling.base/): Base classes for MAX pipeline model definitions.
- [max.pipelines.modeling.dataprocessing](https://docs.modular.com/api/python/pipelines.modeling.dataprocessing/): | [`PaddingDirection`](generated/max.pipelines.modeling.dataprocessing.PaddingDirection.md#max.pipelines.modeling.dataprocessing.PaddingDirection) ...
- [max.pipelines.modeling.types](https://docs.modular.com/api/python/pipelines.modeling.types/): Pipeline modeling types: request/input/pipeline interfaces.
- [max.pipelines.modeling.types.pipeline_variants](https://docs.modular.com/api/python/pipelines.modeling.types.pipeline_variants/): | [`BatchType`](generated/max.pipelines.modeling.types.pipeline_variants.BatchType.md#max.pipelines.modeling.types.pipeline_variants.BatchType)    ...
- [max.pipelines.request](https://docs.modular.com/api/python/pipelines.request/): Request types for MAX API.
- [max.pipelines.request.provider_options](https://docs.modular.com/api/python/pipelines.request.provider_options/): Provider-specific options for MAX platform and modalities.
- [max.pipelines.sampling](https://docs.modular.com/api/python/pipelines.sampling/): | [`SamplingConfig`](generated/max.pipelines.sampling.SamplingConfig.md#max.pipelines.sampling.SamplingConfig) | Configuration for the sampling sta...
- [max.pipelines.speculative](https://docs.modular.com/api/python/pipelines.speculative/): Speculative decoding pipelines and configuration for MAX.
- [max.pipelines.weights](https://docs.modular.com/api/python/pipelines.weights/): Weight loading utilities for MAX pipelines.
- [max.profiler.cpu](https://docs.modular.com/api/python/profiler.cpu/): CPU diagnostics API.
- [max.profiler.gpu](https://docs.modular.com/api/python/profiler.gpu/): GPU diagnostics API.
- [max.profiler](https://docs.modular.com/api/python/profiler/): Performance profiling and tracing utilities for MAX.
- [max.support.image](https://docs.modular.com/api/python/support.image/): Image-related utilities.