Downloads · 30 days
118
7% of all-time downloads
DeepSeek-V3.2-345B-W3A16
0xSero/DeepSeek-V3.2-345B-W3A16
DeepSeek-V3.2-345B-W3A16 is a text generation model from 0xSero. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
[!TIP] Support this work → · X · GitHub · REAP paper · Cerebras REAP
All-time downloads
1.7K
Public
Parameters
1.9B
138 GB on disk
Likes
10
Public
Repo makeup
Click a slice to open those files.
.safetensors138 GB · 100%
Parameter types
How the weights are stored.
I3232.3B · 88%
Try a prompt
Base models
Model card
From the Hugging Face model README
[!TIP] Support this work → · X · GitHub · REAP paper · Cerebras REAP
DeepSeek-V3.2-345B-W3A16
W3A16 quantization of deepseek-ai/DeepSeek-V3.
At a glance
| Base model | deepseek-ai/DeepSeek-V3 |
| Format | W3A16 |
| Total params | 345B |
| Active / token | — |
| Experts / layer | 128 |
| Layers | 61 |
| Hidden size | 7168 |
| Context | 163,840 |
| On-disk size | 138 GB |
Which variant should I pick?
<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> </p>DeepSeek-V3.2-REAP-345B-W3A16
REAP-pruned + W3A16 quantized DeepSeek-V3.2 for efficient deployment.
📋 Model Specifications
| Property | Value |
|---|---|
| Base Model | DeepSeek-V3.2 |
| Parameters | 345B |
| Quantization | W3A16 (3-bit weights) |
🔬 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
License & citation
License inherited from the base model.
@misc{lasby2025reap,
title = {REAP the Experts: Why Pruning Prevails for One-Shot MoE Compression},
author = {Mike Lasby and Ivan Lazarevich and Nish Sinnadurai and Sean Lie and Yani Ioannou and Vithursan Thangarasa},
year = {2025}, eprint = {2510.13999}, archivePrefix = {arXiv}
}
Sponsors
Made possible by NVIDIA · TNG Technology · Lambda · Prime Intellect · Hot Aisle.
- Task
- Text Generation
- Library
- transformers
- Type
- deepseek_v3
- License
- apache-2.0
- Languages
- en
- Created
- Jan 3, 2026
- Updated
- May 30, 2026