Step-3.5-Flash-REAP-121B-A11B
✨ Highlights
Introducing Step-3.5-Flash-REAP-121B-A11B, a memory-efficient compressed variant of Step-3.5-Flash that maintains near-identical performance while being 40% lighter.
This model was created using REAP (Router-weighted Expert Activation Pruning), a novel expert pruning method that selectively removes redundant experts while preserving the router's independent control over remaining experts. Key features include:
- Near-Lossless Performance: Maintains almost identical accuracy on code generation, agentic coding, and function calling tasks compared to the full 196B model
- 40% Memory Reduction: Compressed from 196B to 121B parameters, significantly lowering deployment costs and memory requirements
- Preserved Capabilities: Retains all core functionalities including code generation, math & reasoning and tool calling.
- Drop-in Compatibility: Works with vanilla vLLM - no source modifications or custom patches required
- Optimized for Real-World Use: Particularly effective for resource-constrained environments, local deployments, and academic research
📋 Model Overview
Step-3.5-Flash-REAP-121B-A11B has the following specifications:
- Base Model: Step-3.5-Flash
- Compression Method: REAP (Router-weighted Expert Activation Pruning)
- Compression Ratio: 40% expert pruning
- Type: Sparse Mixture-of-Experts (SMoE) Causal Language Model
- Number of Parameters: 121B total, 11B activated per token
- Number of Layers: 45
- Number of Attention Heads: 64
- Number of Experts: 173 (uniformly pruned from 288)
- Number of Activated Experts: 8 per token
- Context Length: 262,144 tokens
- License: Apache 2.0
📊 Evaluations
🚀 Deployment
You can deploy the model directly using the latest vLLM (that supports Step-3.5-Flash), no source modifications or custom patches required.
vllm serve cerebras/Step-3.5-Flash-REAP-121B-A11B \
--tensor-parallel-size 8 \
--tool-call-parser step3p5 \
--reasoning-parser step3p5 \
--trust-remote-code \
--enable_expert_parallel \
--disable-cascade-attn \
--enable-auto-tool-choice
If you encounter insufficient memory when running this model, you might need to set a lower value for --max-num-seqs flag (e.g. set to 64). For more information, refer to the official vLLM deployment guide.
🧩 Model Creation
This checkpoint was created by applying the REAP (Router-weighted Expert Activation Pruning) method uniformly across all Mixture-of-Experts (MoE) blocks of Step-3.5-Flash, with a 40% pruning rate.
How REAP Works
REAP selects experts to prune based on a novel saliency criterion that considers both:
- Router gate values: How frequently and strongly the router activates each expert
- Expert activation norms: The magnitude of each expert's output contributions
This dual consideration ensures that experts contributing minimally to the layer's output are pruned, while preserving those that play critical roles in the model's computations.
Key Advantages
- One-Shot Compression: No fine-tuning required after pruning - the model is immediately ready for deployment
- Preserved Router Control: Unlike expert merging methods, REAP maintains the router's independent, input-dependent control over remaining experts, avoiding "functional subspace collapse"
- Generative Task Superiority: REAP significantly outperforms expert merging approaches on generative benchmarks (code generation, creative writing, mathematical reasoning) while maintaining competitive performance on discriminative tasks
📚 For more details, refer to the following resources:
⚖️ License
This model is derived from
stepfun-ai/Step-3.5-Flash
and distributed under the Apache 2.0 license.
🧾 Citation
If you use this checkpoint, please cite the REAP paper:
@article{lasby-reap,
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}
}