---
title: MiniMax-M3-EAGLE3-GQA
canonical_url: "https://www.modelscope.cn/models/Inferact/MiniMax-M3-EAGLE3-GQA"
md_url: "https://www.modelscope.cn/models/Inferact/MiniMax-M3-EAGLE3-GQA.md"
repository: Inferact/MiniMax-M3-EAGLE3-GQA
last_updated: 2026-07-15
license: mit
pipeline_tag: text-generation
tasks:
  - text-generation
model_type:
  - llama
architectures:
  - LlamaForCausalLMEagle3
base_model:
  - MiniMaxAI/Minimax-M3-preview
base_model_relation: finetune
parameters: 3.1B
tensor_type:
  - BF16
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 93
stars: 0
tags:
  - eagle3
  - speculative-decoding
  - draft-model
  - gqa
  - vllm
  - torchspec
  - minimax
---

# MiniMax-M3-EAGLE3-GQA

> MiniMax-M3-EAGLE3-GQA - Inferact 在 ModelScope 开源的模型。Inferact/MiniMax-M3-EAGLE3-GQA is a grouped-query-attention (GQA) EAGLE3 draft model for accelerating inference of MiniMax-M3, served with vLLM and trained with TorchSpec.

Inferact/MiniMax-M3-EAGLE3-GQA 是 ModelScope 魔搭社区上的 3.1B 参数text-generation模型，采用 mit 许可，基于 MiniMaxAI/Minimax-M3-preview 构建。

- **Repository**: Inferact/MiniMax-M3-EAGLE3-GQA
- **License**: mit
- **Tasks**: text-generation
- **Parameters**: 3.1B
- **Base model**: MiniMaxAI/Minimax-M3-preview
- **Tags**: eagle3, speculative-decoding, draft-model, gqa, vllm, torchspec, minimax
- **Downloads**: 93
- **Stars**: 0
- **Last updated**: 2026-07-15

Source: https://www.modelscope.cn/models/Inferact/MiniMax-M3-EAGLE3-GQA

---

## Model Overview

**Inferact/MiniMax-M3-EAGLE3-GQA** is a **grouped-query-attention (GQA)** EAGLE3 draft model for accelerating inference of [MiniMax-M3](https://huggingface.co/MiniMaxAI/MiniMax-M3), served with **[vLLM](https://github.com/vllm-project/vllm)** and trained with **[TorchSpec](https://github.com/lightseekorg/TorchSpec)**.

It is **retrained on the same datasets** as the multi-head-attention version [Inferact/MiniMax-M3-EAGLE3](https://huggingface.co/Inferact/MiniMax-M3-EAGLE3) — **kimi-mtp, OpenCodeInstruct, SWE-bench, and SWE-bench-Pro** — with the draft's attention changed from **MHA to GQA** (`num_key_value_heads: 64 → 4`) for **inference efficiency** (16× smaller draft KV cache) and **compatibility with the target model**.

The draft is a 1-layer dense Llama (`LlamaForCausalLMEagle3`) on MiniMax-M3's `hidden_size=6144` / `vocab_size=200064`; at serve time it shares the target's embedding and LM head (EAGLE3). See `config.json` for the full architecture.

---

## Performance

Mean accepted length and draft accept rate measured end-to-end against `MiniMaxAI/MiniMax-M3-MXFP8` served with vLLM at `tensor-parallel-size=4`, `num_speculative_tokens=3`, greedy sampling (`temperature=0`, `top_p=1.0`), `max-concurrency=16`.

| Dataset | n | Mean accepted length | Draft accept rate | Per-position accept rate (pos 1 / 2 / 3) |
|---|---:|---:|---:|---:|
| MT-Bench | 64 | 2.668 | 55.62% | 0.745 / 0.537 / 0.387 |
| SPEED-Bench (qualitative) | 64 | 2.561 | 52.04% | 0.719 / 0.500 / 0.342 |