ConicCat/Qwen3.5-27B-Writer-V2

A tentative second version. Hopefully, it's better.

A writing & roleplay finetune of Qwen3.5 27B. The primary emphasis is on writing quality as it strongly generalizes across both domains.

The basic idea is to use a curriculum learning setup to overcome the lack of high quality roleplay data by first training on lower quality roleplay data, then training on higher quality writing data. Starting from ConicCat/Qwen3.5-Antirep-27B, the model was trained on a roughly equal mixture of instruct / roleplay / writing data for three epochs. The model was then trained for eleven epochs on a smaller dataset of book chunks.

Recommended Settings

• Chatml template with <think>\n\n</think>\n prefill or <think>\n prefill. Should think less!
• temperature = 0.7
• top_p = 0.95
• A moderate dry penalty of ~ 0.4-0.8 should work well.
• For quants, Q4_K_M runs well with ~100k context on 24GB Vram
• IQ4_XS should fit on 16GB Vram with about 20-24k context with the vulkan backend, although it's pretty tight and may require some fiddling around with open programs e.t.c.

Datasets

• ConicCat/AntiRep to mitigate repetitition.

• internlm/Condor-SFT-20K for instruct; even though instruct capabilities are not the primary focus, adding some instruct data helps mitigate forgetting and maintains general intellect and instruction following capabilites.

• ConicCat/Gutenberg-SFT. A reformatted version of the original Gutenberg DPO dataset by jondurbin for SFT with some slight augmentation to address many of the samples being overly long.

• ConicCat/MiniC2_V3.2. The venerable C2, with cleaned and reformatted system prompts, and all user / assistant turns replaced by V3.2.

• A dataset of backtranslated books. Unfortunately, I am unable to release this set as all of the data is under copyright.