jackyk02/tulu3-pref-qwen3.6-35b-a3b-fp8-embeddings-256k

Tulu 3 preference mixture: GPT-4o-rated pairs, the rejected response as hard negative: Qwen3.6-35B-A3B-FP8 state/action embeddings (state, action) embedding pairs from Qwen/Qwen3.6-35B-A3B-FP8 (vLLM pooling runner, last-token pooling, L2-normalized, 2048-d) for post-training contrastive (CLM) heads, built with clm_train (docs/tulu3-pref.md, DATASET=tulu3-pref scripts/embed_chat_sft.sbatch). Sources: allenai/llama-3.1-tulu-3-405b-preference-mixture split pairs tasks pairs… See the full description on the dataset page: https://huggingface.co/datasets/jackyk02/tulu3-pref-qwen3.6-35b-a3b-fp8-embed

<iframe src="https://silytics.si/embed/item/e6fe40ff-f8d4-459a-b06f-5e3b0b5667a2" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 24 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.