sahilmob/gpt-oss-20b-toolcall-phase1-v2-strict-lora

gpt-oss-20b-toolcall-phase1-v2-strict-lora

Phase-1 LoRA adapter for Wish Engine implementor tool-calling.

Training Summary

• —Base model: `openai/gpt-oss-20b`
• —Dataset: `sahilmob/wish-engine-toolcall-next-v2-strict`
• —Method: SFT with LoRA
• —Purpose: increase tool-call format/sequence correctness before broader curriculum phase

Dataset Scope

Strict subset focused on high-signal trajectories:

• —passed runs only
• —JSON tool arguments
• —stronger tool-use depth threshold

Notes

• —This is an intermediate checkpoint used for phase-2 continuation.
• —Final benchmark results should be reported on the final phase-2 model card.
• —Central runbook: `sahilmob/wish-engine-toolcall-runbook`

Limitations

• —Not intended as a general-purpose coding model.
• —Eval metrics are pending publication after phase-2 run completion.