DeepSeek-V4.1-Flash-1017 Coder GGUF model released
Original title: ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-Coder-GGUF — new model trending #30 on Hugging Face
A 763B-parameter MoE model pruned to 72.1 GB and 1.05 bits per parameter, optimized for code and agentic workflows.
ISTA-DASLab released DeepSeek-V4.1-Flash-1017-GSQ-RCO-Coder-GGUF, a pruned version of DeepSeek-V4.1-Flash-1017. Half of the 384 routed experts per layer were removed and the remaining weights are quantized to 1.05 bits per parameter. The model is 72.1 GB and is built for code, agentic tool use, long context and vision. It retains 94.8% of SWE-bench Verified and 98.7% of LiveCodeBench v6 performance. This is an experimental release.
Unconfirmed: Performance on tasks outside code and agentic use is not reported
- 763B MoE model compressed to 72.1 GB
- 1.05 bits per parameter effective rate
- 192 of 384 experts removed per layer
- 94.8% SWE-bench retention
- 98.7% LiveCodeBench retention
- Vendors
- ISTA-DASLab
- Products
- DeepSeek-V4.1-Flash-1017
- Organizations
- IST-DASLab
From the original article
8 min read at huggingface.coimage-text-to-text · 73,282 downloads · 156 likes · license mit
By ISTA-DASLab · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.