𓌳 REAP𓌳 the Experts: Why Pruning Prevails for One-Shot MoE Compression
📄 Paper • 💻 Code

GLM-4.6-REAP-218B-A32B-W4A16-AutoRound

W4A16 quantized version of Cerebras' official GLM-4.6-REAP-218B-A32B.

• ~4x Size Reduction: ~436GB → ~116GB
• Runs on Consumer Hardware: 8x RTX 3090 or 4x RTX 4090
• vLLM/SGLang Compatible: Drop-in deployment

🙏 Acknowledgments

• Prime Intellect — Compute sponsorship
• Cerebras — REAP base model
• Intel — AutoRound quantization

---

📋 Model Specifications

Property	Value
Base Model	cerebras/GLM-4.6-REAP-218B-A32B
Parameters	218B total, 32B activated
Quantization	W4A16 (4-bit weights, 16-bit activations)
Original Size	~436GB
Quantized Size	~116GB

---

📊 Benchmarks

Tested on 8x RTX 3090:

Metric	Value
Prompt Tokens	~21,178
Completion Tokens	393
Time to First Token	23.82s

---

🚀 Deployment

vLLM

vllm serve 0xSero/GLM-4.6-REAP-218B-A32B-W4A16-AutoRound \
    --tensor-parallel-size 8 \
    --trust-remote-code \
    --quantization gptq

---

🔬 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	700	Code generation	51% of mix — Code tasks activate specific expert pathways; pruning without code calibration destroys coding ability
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	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

---

🧾 Citation

@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}
}