Skip to main content
Ctrl+K
MaxText  documentation - Home MaxText  documentation - Home
  • MaxText
  • Install MaxText
  • Build and Upload MaxText Docker Images
  • Tutorials
    • Getting started: First run
    • Pre-training
    • Post-training
      • SFT on single-host TPUs
      • SFT on multi-host TPUs
      • Reinforcement Learning on single-host TPUs
      • Reinforcement Learning on Multi-Host TPUs
      • Knowledge distillation
      • Multimodal support
      • Full fine-tuning on single-host TPUs
      • GEPA Prompt Optimization for MaxText
    • Inference on MaxText
  • Run MaxText
    • Via localhost or single-host VM
    • Via single-host GPU
    • At scale with XPK
    • Via Pathways
    • Via Decoupled Mode (No Google Cloud Dependencies)
  • How-to guides
    • Optimization
      • Customizing model configs for TPUs
      • Sharding on TPUs
      • Optimizing with Pallas kernels
      • Benchmarking & tuning guide
    • Data pipelines
      • Grain pipeline
      • Hugging Face pipeline
      • TFDS pipeline
      • Optimizing pipeline performance
    • Checkpointing
      • GCS bucket-based checkpointing
      • Emergency checkpointing
      • Multi-tier checkpointing
      • Checkpoint Conversion Utilities
    • Monitoring and debugging
      • Features and diagnostics
      • Enable GCP workload observabiltiy
      • Troubleshooting: Megascale hangs
      • ML Goodput measurement
      • Understand logs and metrics
      • Use Vertex AI Tensorboard
      • Profiling with XProf
      • Running a workload with Google Cloud ML Diagnostics Enabled
    • Run MaxText Python Notebooks on TPUs
    • MaxText Model Bringup: Community Contributor Guide
    • Distillation
  • Reference documentation
    • Performance metrics
    • Models
      • Optimized models tiering
      • Supported models list
    • Architecture
      • Architecture overview
      • MaxText and the JAX ecosystem
    • Core concepts
      • Checkpoints
      • Comparison to alternatives
      • Batch Size
      • Quantization
      • Tiling
      • JAX, XLA, and Pallas
      • Mixture of Experts (MoE) Configuration
  • How to Contribute
    • Update MaxText dependencies
    • Contribute to documentation
  • MaxText release notes
  • .rst

maxtext.integration.tunix package

Contents

  • Subpackages
  • Submodules

maxtext.integration.tunix package#

Subpackages#

  • maxtext.integration.tunix.weight_mapping package
    • StandaloneVllmWeightMapping
    • Submodules
      • maxtext.integration.tunix.weight_mapping.deepseek3 module
        • DEEPSEEK_VLLM_MAPPING
      • maxtext.integration.tunix.weight_mapping.gpt_oss module
        • GPT_OSS_VLLM_MAPPING
      • maxtext.integration.tunix.weight_mapping.llama3 module
        • LLAMA3_VLLM_MAPPING
      • maxtext.integration.tunix.weight_mapping.qwen2 module
        • QWEN2_VLLM_MAPPING
      • maxtext.integration.tunix.weight_mapping.qwen3 module
        • QWEN3_VLLM_MAPPING

Submodules#

  • maxtext.integration.tunix.tunix_adapter module
    • TunixMaxTextAdapter
      • TunixMaxTextAdapter.to_hf_mappings()
      • TunixMaxTextAdapter.to_hf_transpose_keys()
      • TunixMaxTextAdapter.to_hf_hook_fns()
      • TunixMaxTextAdapter.lora_to_hf_mappings()
  • maxtext.integration.tunix.utils module
    • VllmWeightMapping
      • VllmWeightMapping.to_hf_mapping()
      • VllmWeightMapping.to_hf_transpose_keys()
      • VllmWeightMapping.to_hf_hook_fns()
      • VllmWeightMapping.lora_to_hf_mappings()
      • VllmWeightMapping.convert_hf_map_to_sharding_map()
Contents
  • Subpackages
  • Submodules

By MaxText developers

© Copyright 2023–2026, Google LLC.