ciocan/ornith-1.5-9b-spreadsheetbench-merged

Ornith 1.5 9B, SpreadsheetBench LoRA fine-tune, merged bf16 weights

The merged weights of ciocan/ornith-1.5-9b-spreadsheetbench-lora applied to ornith-ai/Ornith-1.5-9B, as served for Ambit Labs' Ylookup x Encode hackathon entry. Harness, data recipe, evaluation and write-up: https://github.com/Ambit-Labs/hacks-enyl-research.

Serve with vLLM 0.28:

vllm serve <this repo> --served-model-name Ornith-9B-ft \
  --tool-call-parser qwen3_xml --reasoning-parser qwen3 --enable-auto-tool-choice \
  --max-model-len 32768 --trust-remote-code

Use chat_template_kwargs: {"enable_thinking": false} for structured tool calls at temperature 0. Results and caveats are on the adapter's model card.