Llama3.2 math-retention pilot
Built with Llama. Experimental adapter; not validated mathematical reasoning.
Load from resume at the selected immutable commit in the returned CPU report. Latest commit may be a rejected development checkpoint.
Upstream revision: 0cb88a4f764b7a12671c53f0838cd831a0843b95
Historical tokenizer: {"repo":"fc91/CQoT_SFT_mix_teacher_repair_round3_Llama-3.2-3B-HPC","revision":"1c0a3554332b7adaa03edf51df7840aa2b7652f4"}
256 checked direct-solution targets. Math-token CE plus forward reference KL. Public checkpoints include optimizer/RNG for recovery. No held-out responses are uploaded.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for fc91/Llama-3.2-3B_CQoT_MathRetention_20261007_public
Base model
meta-llama/Llama-3.2-3B-Instruct