⚡ Qwopus3.8-27B-Flash EXL3
Follow me on X @BennyDaBall_OG!
EXL3 quants of Jackrong/Qwopus3.8-27B-Flash, with the model's MTP draft head and vision tower retained.
📦 Available quants
| Branch | Decoder | LM head | MTP | Vision |
|---|---|---|---|---|
| 3.50bpw | 3.50 bpw | 6 bpw | 4 bpw | BF16 |
Each weight branch contains its own model card, exact conversion recipe, runtime notes, tokenizer, chat template, and multimodal processor files.
hf download BennyDaBall/Qwopus3.8-27B-Flash-EXL3 \
--revision 3.50bpw \
--local-dir Qwopus3.8-27B-Flash-EXL3-3.50bpw
Built for ExLlamaV3 and TabbyAPI. Apache-2.0, matching the upstream model metadata.
🙏 Credits
- Jackrong for Qwopus3.8-27B-Flash.
- Qwen for Qwen3.8-27B.
- turboderp-org/exllamav3 and theroyallab/tabbyAPI.