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

GLM-4.7-REAP-40

✨ Highlights

40% Expert-Pruned GLM-4.7 optimized for code generation, function calling, and agentic workflows.

Created using REAP (Router-weighted Expert Activation Pruning) by Cerebras:

• 358B → 218B: 40% of MoE experts pruned (96/160 remaining)
• Calibrated for Code & Tools: Preserves coding and function-calling capabilities
• One-Shot Compression: No fine-tuning required
• Drop-in Compatible: Works with vLLM, Transformers, SGLang

🙏 Acknowledgments

• Prime Intellect — Compute sponsorship (8x H200 cluster)
• Cerebras — REAP methodology

---

📋 Model Specifications

Property	Value
Base Model	zai/glm-4.7
Architecture	Sparse Mixture-of-Experts (SMoE)
Original Parameters	358B
Pruned Parameters	218B
Compression	40% experts removed
Experts per Layer	96 (was 160)
MoE Layers	92
Activated Experts	8 per token
Precision	BF16
Disk Size	~407GB
VRAM Required	~407GB

---

🔬 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

---

📦 Related Models

Model	Params	Experts	Size	Format
GLM-4.7-REAP-30	251B	112	~470GB	BF16
GLM-4.7-REAP-35	233B	104	~439GB	BF16
GLM-4.7-REAP-40	218B	96	~407GB	BF16
GLM-4.7-REAP-45	197B	88	~370GB	BF16
GLM-4.7-REAP-50	179B	80	~345GB	BF16
GLM-4.7-REAP-40-W4A16	218B	96	~108GB	GPTQ
GLM-4.7-REAP-50-W4A16	179B	80	~92GB	GPTQ

---

🚀 Deployment

vLLM (Recommended)

vllm serve 0xSero/GLM-4.7-REAP-40 \
    --tensor-parallel-size 8 \
    --trust-remote-code \
    --dtype bfloat16

Transformers

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "0xSero/GLM-4.7-REAP-40",
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("0xSero/GLM-4.7-REAP-40", trust_remote_code=True)

messages = [{"role": "user", "content": "Write a Python function to merge two sorted lists."}]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True)
outputs = model.generate(inputs.to(model.device), max_new_tokens=512, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

🧩 Reproduction

REAP Pruning Script

#!/usr/bin/env python3
"""
REAP Pruning Script for MoE Models
Adapted from: https://github.com/CerebrasResearch/reap
"""

import subprocess
import sys

def run_reap(
    model_path: str,
    compression_ratio: float,
    dataset: str = "0xSero/glm47-reap-calibration-v2",
    samples: int = 1360,
    seed: int = 42,
    distance: str = "angular",
    reuse_observations: str = None,
):
    """
    Run REAP expert pruning.

    Args:
        model_path: Path to base model
        compression_ratio: 0.30 = prune 30%, keep 70%
        dataset: Calibration dataset (code + tools + agentic)
        samples: Number of calibration samples
        seed: Random seed for reproducibility
        distance: Distance metric for expert clustering
        reuse_observations: Path to pre-computed observations for instant pruning
    """
    cmd = [
        sys.executable, "src/reap/prune.py",
        "--model-name", model_path,
        "--dataset-name", dataset,
        "--compression-ratio", str(compression_ratio),
        "--prune-method", "reap",
        "--seed", str(seed),
        "--samples_per_category", str(samples),
        "--model_max_length", "2048",
        "--distance_measure", distance,
        "--record_pruning_metrics_only", "true",
    ]

    if reuse_observations:
        # Instant pruning: skip calibration, reuse precomputed expert scores
        cmd.extend(["--load_observations", reuse_observations])

    subprocess.run(cmd, check=True)

# Example: Create 40% pruned model
run_reap(
    model_path="/path/to/GLM-4.7",
    compression_ratio=0.40,  # Prune 40% of experts
)

Observation Reuse (Instant Multi-Ratio Pruning)

REAP computes expert saliency scores during calibration. These scores are compression-ratio independent, enabling instant pruning at any ratio:

# First run: compute observations (~5 hours)
python prune.py --compression-ratio 0.40 --output_file_name observations.pt

# Subsequent runs: instant pruning (<5 minutes)
python prune.py --compression-ratio 0.30 --load_observations observations.pt
python prune.py --compression-ratio 0.50 --load_observations observations.pt

---

⚖️ License

Apache 2.0 (inherited from GLM-4)

---

🧾 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}
}