---
title: MiniMax-M2.1-REAP-50
canonical_url: "https://www.modelscope.cn/models/0xSero/MiniMax-M2.1-REAP-50"
md_url: "https://www.modelscope.cn/models/0xSero/MiniMax-M2.1-REAP-50.md"
repository: 0xSero/MiniMax-M2.1-REAP-50
last_updated: 2026-01-04
license: apache-2.0
pipeline_tag: text-generation
tasks:
  - text-generation
model_type:
  - minimax_m2
architectures:
  - MiniMaxM2ForCausalLM
base_model:
  - MiniMaxAI/MiniMax-M2.1
base_model_relation: quantized
parameters: 81.4B
tensor_type:
  - BF16
  - F8_E4M3
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
language:
  - en
downloads: 40
stars: 1
tags:
  - minimax
  - MOE
  - pruning
  - compression
  - reap
  - cerebras
  - code
  - function-calling
---

# MiniMax-M2.1-REAP-50

> MiniMax-M2.1-REAP-50 - 0xSero 在 ModelScope 开源的模型。𓌳 REAP 𓌳 the Experts: Why Pruning Prevails for One-Shot MoE Compression 📄 Paper • 💻 Code • 📝 Blog

0xSero/MiniMax-M2.1-REAP-50 是 ModelScope 魔搭社区上的 81.4B 参数text-generation模型，采用 apache-2.0 许可，基于 MiniMaxAI/MiniMax-M2.1 构建。

- **Repository**: 0xSero/MiniMax-M2.1-REAP-50
- **License**: apache-2.0
- **Tasks**: text-generation
- **Parameters**: 81.4B
- **Base model**: MiniMaxAI/MiniMax-M2.1
- **Tags**: minimax, MOE, pruning, compression, reap, cerebras, code, function-calling
- **Downloads**: 40
- **Stars**: 1
- **Last updated**: 2026-01-04

Source: https://www.modelscope.cn/models/0xSero/MiniMax-M2.1-REAP-50

---

<p align="center">
  <em>𓌳 <strong>REAP</strong>𓌳  the Experts: Why Pruning Prevails for One-Shot MoE Compression</em><br>
  <a href="https://arxiv.org/abs/2510.13999">📄 Paper</a> • <a href="https://github.com/CerebrasResearch/reap">💻 Code</a> • <a href="https://www.cerebras.ai/blog/reap">📝 Blog</a>
</p>

# MiniMax-M2.1-REAP-50

## ✨ Highlights

**50% Expert-Pruned** MiniMax-M2.1 optimized for **code generation** and **function calling**.

- **50% Expert Pruning**: ~~80 experts remaining per layer
- **Calibrated for Code & Tools**: Same calibration mix as GLM-4.7 REAP models
- **One-Shot Compression**: No fine-tuning required

### 🙏 Acknowledgments

- **[Prime Intellect](https://www.primeintellect.ai/)** — Compute sponsorship (8x H200 cluster)
- **[Cerebras](https://www.cerebras.net/)** — [REAP methodology](https://arxiv.org/abs/2510.13999)

---

## 📋 Model Specifications

| Property | Value |
|----------|-------|
| **Base Model** | MiniMax-M2.1 |
| **Compression** | 50% experts removed |
| **Parameters** | ~220B |
| **Experts per Layer** | ~80 |
| **Precision** | BF16 |
| **Disk Size** | ~420GB |

---


## 🔬 Calibration Dataset: Deep Dive

REAP's effectiveness depends critically on **calibration data that represents the target use case**. We specifically optimized for **code generation**, **function/tool calling**, and **agentic workflows**.

### Why These 3 Datasets?

| Dataset | Samples | Purpose | Why It Matters |
|---------|---------|---------|----------------|
| [evol-codealpaca-v1](https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1) | 700 | Code generation | **51% of mix** — Code tasks activate specific expert pathways; pruning without code calibration destroys coding ability |
| [xlam-function-calling-60k](https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k) | 330 | Function/tool calling | **24% of mix** — Tool use requires structured JSON output; experts handling schema generation must be preserved |
| [SWE-smith-trajectories](https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories) | 330 | Agentic multi-turn | **24% of mix** — Real SWE-bench trajectories with tool calls, file edits, and multi-step reasoning |

### The Science Behind Dataset Selection

```
REAP Algorithm:
1. Forward pass calibration samples through model
2. Record which experts activate and their magnitudes
3. Compute saliency = router_weight × activation_norm
4. Prune lowest-saliency experts

Key Insight: Experts are TASK-SPECIFIC
├── Some experts specialize in natural language
├── Some experts specialize in code syntax
├── Some experts specialize in JSON/structured output
└── Some experts specialize in multi-turn context

If calibration lacks code → code-specialized experts appear "unused" → get pruned → model loses coding ability
```

### Cerebras' Original Mix (from paper)

Cerebras used the same 3 datasets in their GLM-4.6 REAP experiments:
- evol-codealpaca-v1 for code generation
- xlam-function-calling-60k for tool calling
- SWE-smith-trajectories for agentic tasks

We followed this exact recipe for reproducibility.

### Combined Dataset

Our calibration mix: [0xSero/glm47-reap-calibration-v2](https://huggingface.co/datasets/0xSero/glm47-reap-calibration-v2)


---

## 📦 Related Models

| Model | Compression | Experts | Size |
|-------|-------------|---------|------|
| [MiniMax-M2.1-REAP-25](https://huggingface.co/0xSero/MiniMax-M2.1-REAP-25) | 25% | ~120 | ~620GB |
| [MiniMax-M2.1-REAP-30](https://huggingface.co/0xSero/MiniMax-M2.1-REAP-30) | 30% | ~112 | ~580GB |
| [MiniMax-M2.1-REAP-40](https://huggingface.co/0xSero/MiniMax-M2.1-REAP-40) | 40% | ~96 | ~500GB |
| [MiniMax-M2.1-REAP-50](https://huggingface.co/0xSero/MiniMax-M2.1-REAP-50) | 50% | ~80 | ~420GB |

---

## 🚀 Deployment

### vLLM

```bash
vllm serve 0xSero/MiniMax-M2.1-REAP-50 \
    --tensor-parallel-size 8 \
    --trust-remote-code \
    --dtype bfloat16
```

---

## 🧩 Reproduction

### REAP Pruning

```bash
#!/bin/bash
# MiniMax REAP - same calibration as GLM-4.7

export MODEL_DIR=/path/to/MiniMax-M2.1
export REAP_DATASET=0xSero/glm47-reap-calibration-mix
export REAP_SAMPLES_PER_CATEGORY=999
export REAP_MODEL_MAX_LENGTH=2048

python src/reap/prune.py \
    --model-name $MODEL_DIR \
    --dataset-name $REAP_DATASET \
    --compression-ratio 0.50 \
    --prune-method reap \
    --seed 42 \
    --distance_measure cosine
```

---

## ⚖️ License

Apache 2.0

---

## 🧾 Citation

```bibtex
@article{lasby2025reap,
  title={REAP the Experts: Why Pruning Prevails for One-Shot MoE Compression},
  author={Lasby, Mike and Lazarevich, Ivan and Sinnadurai, Nish and Lie, Sean and Ioannou, Yani and Thangarasa, Vithursan},
  journal={arXiv preprint arXiv:2510.13999},
  year={2025},
  url={https://arxiv.org/abs/2510.13999}
}
```