Sign inGet started
Dataset · Hugging Face

clm-project/jev-boost-kn2-acecode-output-qwen3.6-35b-a3b-fp8-embeddings

Jev boost: kn2-acecode-output: Qwen3.6-35B-A3B-FP8 state/action embeddings TIGER-Lab/AceCode-87K (revision 13216309a9f6cb40b60cb1a9750071efeac414ad), split train, converted into decisions shaped like the Jev Decision Index's Knowledge & Reasoning requests (CRUXEval) and embedded with Qwen/Qwen3.6-35B-A3B-FP8 (vLLM pooling, last token, L2-normalized, 2048-d) for post-training CLM heads (clm_train: clm_train.data.jev_boost, docs/jev-boost.md). a passing solution's output; failing… See the full description on the dataset page: https://huggingface.co/datasets/clm-project/jev-boost-kn2-acecode-outp

Author / organizationclm-project
License (as declared)other
Downloads0 (as reported by Hugging Face)
Likes0
Created25 October 2026
Last updated at source25 October 2026
embeddingscontrastive-learningqwen3.6jev-decision-index
Embed this entry
<iframe src="https://silytics.si/embed/item/591bc68c-7533-4275-930a-eef91e713c23" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Free, no tracking. More about feeds and embeds.

More from clm-project

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 25 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.