OLMo BFCL r32 tool-call artifacts
Artifacts for the krishna/olmo-bfcl-recovery run on issue #2.
Base model: allenai/OLMo-7B-0724-Instruct-hf.
Training data: strict10k BFCL mix, preserving existing <tool_call>...</tool_call> target text.
Adapter rank: 32.
contents
final_adapter/: final 1250-step LoRA adapter.checkpoint_step_000625/: midpoint LoRA adapter.evals/: BFCL eval generations and summaries.logs/: train/eval logs.
scores
| artifact | raw exact | normalized exact |
|---|---|---|
| base instruct OLMo, no adapter | 247/1007 | 253/1007 |
| checkpoint_step_000625 | 0/1007 | 0/1007 |
| final_adapter | 0/1007 | 0/1007 |
The r32 adapter outputs were malformed/repetitive, so the zero score is not just a parser miss.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for Occupying-Mars/olmo-bfcl-r32-tool-call-artifacts
Base model
allenai/OLMo-7B-0724-Instruct-hf