> This page is for version Nightly (default).
> For other versions, use one of these documentation indexes:
> - Nightly (default): https://docs.nvidia.com/nemo/automodel/nightly/llms.txt
> - Latest: https://docs.nvidia.com/nemo/automodel/latest/llms.txt
> - 0.7.0 · 26.10: https://docs.nvidia.com/nemo/automodel/v0.7/llms.txt
> - 0.5.0 · 26.06: https://docs.nvidia.com/nemo/automodel/v0.5/llms.txt
> - 0.4.0 · 26.04: https://docs.nvidia.com/nemo/automodel/v0.4/llms.txt

> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/nemo/automodel/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/nemo/automodel/_mcp/server.

# MiniMax-M3.1-Flash

> Fine-tune MiniMax-M3.1-Flash on MedPix vision-language data with NeMo AutoModel, loading its NVFP4 and MXFP8 checkpoint directly.

[MiniMaxAI/MiniMax-M3.1-Flash](https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash) is an update of [MiniMax-M3](/nemo/automodel/model-coverage/vision-language-models/MiniMaxAI/MiniMax-M3), MiniMax AI's Mixture-of-Experts vision-language model, optimized for agentic coding. NeMo AutoModel loads its mixed-precision checkpoint directly and fine-tunes it with the same model implementation as MiniMax-M3.

## Model Reference

### Model Architecture

| Property                 | Value                                                                                  |
| ------------------------ | -------------------------------------------------------------------------------------- |
| **Task**                 | Image-Text-to-Text                                                                     |
| **Architecture**         | `MiniMaxM3SparseForConditionalGeneration` - MoE VLM, same implementation as MiniMax-M3 |
| **Language Module**      | 60 layers, routed experts, MiniMax Sparse Attention (MSA)                              |
| **Checkpoint Precision** | NVFP4 routed experts, MXFP8 projections, BF16 elsewhere                                |
| **HF Org**               | [MiniMaxAI](https://huggingface.co/MiniMaxAI)                                          |

## Checkpoint Loading

The released checkpoint stores routed-expert weights in packed NVFP4 (values, block scales, and global scales) and attention and indexer projections in MXFP8. With `checkpoint.dequantize_base_checkpoint: true`, AutoModel decodes these tensors into the training dtype while streaming them into sharded storage, so you do not need to convert the checkpoint first.

The recipe trains with BF16 parameters, FP32 router state, and Transformer Engine FusedAdam with FP32 master weights and moments. This is BF16 fine-tuning of the dequantized model, not NVFP4 training.

## Available Models

* **MiniMax-M3.1-Flash** - registered through the `MiniMaxM3SparseForConditionalGeneration` model class.

## Example HF Models

| Model              | HF ID                                                                                 |
| ------------------ | ------------------------------------------------------------------------------------- |
| MiniMax-M3.1-Flash | [`MiniMaxAI/MiniMax-M3.1-Flash`](https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash) |

## Example Recipes

| Recipe                                                                                                                                                         | Scope                                                  | Hardware           | Steps |
| -------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------ | ------------------ | ----- |
| [Full SFT - MedPix, EP128](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/vlm_finetune/minimax_m3/minimax_m3_1_vl_sft_medpix_ep128_100steps.yaml) | All 60 language layers; frozen vision and audio towers | 16 nodes, 128 GPUs | 100   |

The recipe uses FSDP2 with HybridEP expert dispatch, FA4 attention, and activation checkpointing (TP1, CP1, PP1), and runs final validation on `validation[:256]`. Checkpoint saving is disabled.

The processor sets `size.shortest_edge: 3136` and `size.longest_edge: 451584`. These keep the checkpoint's image pixel-area bounds under Transformers 5.18 and are not a fixed image width and height.

## Run the Recipe

The recipe needs 16 nodes with 8 GPUs each. See [SLURM Cluster](/nemo/automodel/job-launchers/slurm-cluster) to submit it as a multi-node job. To log to Weights & Biases, set `wandb.enable: true` and fill in `wandb.project` and `wandb.entity`.

## Hugging Face Model Cards

* [MiniMaxAI/MiniMax-M3.1-Flash](https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash)