Downloads · 30 days
28
0% of all-time downloads
jiangchengchengNLP/Llama-4-Scout-17B-16E-Instruct-abliterated-v2-nvfp4
Llama-4-Scout-17B-16E-Instruct-abliterated-v2-nvfp4 is a machine learning model from jiangchengchengNLP. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Llama-4-Scout-17B-16E-Instruct-abliterated-v2-nvfp4 是基于原模型 jiangchengchengNLP/Llama-4-Scout-17B-16E-Instruct-abliterated-v2 通过 llm-compressor 工具进行压缩与优化后的版本。
Downloads · 30 days
28
0% of all-time downloads
All-time downloads
76.9K
Public
Parameters
63.7B
69.7 GB on disk
Likes
3
Public
Click a slice to open those files.
.safetensors69.7 GB · 100%
How the weights are stored.
U851.3B · 81%
From the Hugging Face model README
Llama-4-Scout-17B-16E-Instruct-abliterated-v2-nvfp4 是基于原模型
jiangchengchengNLP/Llama-4-Scout-17B-16E-Instruct-abliterated-v2
通过 llm-compressor 工具进行压缩与优化后的版本。
该版本采用 NVFP4 (NVIDIA FP4) 精度格式与 Mixture-of-Experts (MoE) 架构, 在保持生成质量的同时显著提升了推理速度与显存利用率。
| 项目 | 内容 |
|---|---|
| 基础架构 | Llama 4 |
| 参数规模 | 17B |
| 专家数 | 16 Experts (MoE) |
| 精度 | NVFP4 |
| 优化工具 | llm-compressor |
| 许可证 | Apache 2.0 |
| 原始模型 | jiangchengchengNLP/Llama-4-Scout-17B-16E-Instruct-abliterated-v2 |
| 显卡型号 | 架构 | 是否支持 |
|---|---|---|
| NVIDIA H100 | Hopper | ✅ 支持 |
| NVIDIA B200 | Blackwell | ✅ 支持 |
| NVIDIA RTX 5090 | Blackwell (SM120) | ❌ 暂不支持 |
当前版本的 NVFP4-MoE 算子 暂未在
sm120架构(如 RTX 5090)上适配。
vllm serve jiangchengchengNLP/Llama-4-Scout-17B-16E-Instruct-abliterated-v2-nvfp4 --port 6006 --host 0.0.0.0 --max-model-len 20000 --served-model-name llama4 --gpu-memory-utilization 0.95 --tensor-parallel-size 4
temperature = 0.7
top_p = 0.95
本模型依据 Apache License 2.0 授权,可用于商业用途。