meirdick/router-expert-code.strings_and_lists

1---2base_model: Qwen/Qwen3-4B-Instruct-25073library_name: peft4license: apache-2.05tags:6- lora7- router-expert8---9 10# router-expert-code.strings_and_lists11 12LoRA expert for the `code.strings_and_lists` target of the route-then-admit pool, trained on `meirdick/router-experts-data` `data/experts/code.strings_and_lists.jsonl`.13 14- base: `Qwen/Qwen3-4B-Instruct-2507`15- rank 16, alpha 32, dropout 0.05, modules q_proj, k_proj, v_proj, o_proj16- lr 0.0002, epochs 2, max_len 1024, token budget 8192 per batch17- 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 options18- loss on the assistant turn only; held-out 5% for the val loss19 20| field | value |21|---|---|22| target | code.strings_and_lists |23| items | 1500 |24| recipes | direct,cot_short |25| kinds | {'code': 1500} |26| mc_padded | 0 |27| examples | 3000 |28| encoded | 3000 |29| dropped_too_long | 0 |30| train_rows | 2850 |31| val_rows | 150 |32| train_batches | 55 |33| train_tokens | 440804 |34| supervised_tokens | 97440 |35| steps | 110 |36| train_loss | 0.1857410685413263 |37| val_loss_before | 1.3176491486047464 |38| val_loss | 0.1612570580733553 |39| seconds | 204.85 |40| trainable_params | 11796480 |41