CoolFace
Modelpublic

AzJoeLesz/mathir-parser-qwen25-7b-lora-v3

sourceHugging Faceupdated 4d agoView on Hugging Face
0likes23downloads
Model Card

MathIR parser LoRA v3 — DEPRECATED

Do not integrate. Frozen gold: structural 1/23, status 1/23; status missing on 21/23 preds (schema drift away from MathIR).

Use AzJoeLesz/mathir-parser-qwen25-7b-lora-v2 until a recovered candidate (≥ E4 gold) is scored.

Negative result kept for postmortem; train was from-scratch 400 steps on mathir-synth-hu@v3.