DeepSeek-V4.1-Flash-1017 — IQR IQ3R
Radiance .rad of DeepSeek-V4.1-Flash-1017.
Weights are not in this repo yet. The file is still being written.
Container size: 307.60 GiB (planned).
| Container | 307.60 GiB planned |
| Routed experts | IQ3R on every layer (u3, tcq 16, mad, fwht128, gain 1.08) |
| Attention | u6 / u7 |
| Engram table | iq4_nl |
| Embeddings / output | q6_K |
| Source | deepseek-ai/DeepSeek-V4.1-Flash-1017 (fp8 weights, fp4 experts) plus Engram tables rebuilt from the tokenizer |
Other containers: uncensored UD-IQ1_S, 251.13 GiB · IQR hybrid, 262.75 GiB
Index: https://huggingface.co/collections/bakon3/deepseek-v41-flash-radiance-6af1ddf468e089a2acfccafd