CoolFace
Modelpublic

meirdick/router-expert-code.algorithms

sourceHugging Faceapache-2.0updated 4d agoView on Hugging Face
0likes15downloads
Model Card

router-expert-code.algorithms

LoRA expert for the code.algorithms target of the route-then-admit pool, trained on meirdick/router-experts-data data/experts/code.algorithms.jsonl.

  • base: Qwen/Qwen3-4B-Instruct-2507
  • rank 16, alpha 32, dropout 0.05, modules qproj, kproj, vproj, oproj
  • lr 0.0002, epochs 2, max_len 1024, token budget 8192 per batch
  • one example per item per recipe (direct,cot_short), under the serving system prompt; 40% of the mc items re-lettered to 5 to 10 options
  • loss on the assistant turn only; held-out 5% for the val loss
fieldvalue
targetcode.algorithms
items1500
recipesdirect,cot_short
kinds{'code': 1500}
mc_padded0
examples3000
encoded2998
droppedtoolong2
train_rows2848
val_rows150
train_batches97
train_tokens767364
supervised_tokens387246
steps194
train_loss0.14857557573422944
vallossbefore0.9747351106604415
val_loss0.12078290439546512
seconds269.24
trainable_params11796480