AtomicChat/DeepSeek-V4.1-Flash-1017-Uncensored-GSQ-RCO-GGUF
AI model · https://huggingface.co/AtomicChat/DeepSeek-V4.1-Flash-1017-Uncensored-GSQ-RCO-GGUF
First seen by Oriel on .
AI summary
This Hugging Face repository provides GSQ-RCO GGUF quantizations of DeepSeek-V4.1-Flash-1017 with refusal-direction projections removed while preserving task accuracy and compatibility with Strata.
- Only attn_output_b, ffn_down_shexp, and engram_wkv tensors are replaced; routed experts remain byte-for-byte identical to ISTA-DASLab’s quantization.
- The published IQ2_XXS, IQ2_XS, and IQ1_M rungs reduce JailbreakBench refusal results to 0.0%, with minimal distribution shift.
- On IQ2_XXS, GPQA-Diamond remains 83.3% and AIME25 changes from 76.7% to 80.0%, within the stated benchmark noise.
- A 763B model runs on a single 12 GB GPU using system memory and a separately downloaded 110 GB Engram file streamed from SSD.
- The edit may miss refusals appearing later, removes refusals broadly, and can produce long plans when thinking is enabled.
#deepseek #gguf #model quantization #strata
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Models: 723, latest observation ; +620 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Models: 207,466, latest observation ; +197,748 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | AtomicChat/DeepSeek-V4.1-Flash-1017-Uncensored-GSQ-RCO-GGUF | https://huggingface.co/AtomicChat/DeepSeek-V4.1-Flash-1017-Uncensored-GSQ-RCO-GGUF Read discussion |