| title: "Order the Lab: a tool-calling trace, and what the model said with no tools" | |
| emoji: 🩺 | |
| colorFrom: red | |
| colorTo: gray | |
| sdk: static | |
| pinned: false | |
| # Order the Lab: a tool-calling trace, and what the model said with no tools | |
| MIS 752 Lab 6, Week 6 (Function / Tool Calling). `qwen2.5:3b-instruct` on `ollama` was given `2` Python tools (forecast_ed_arrivals, get_drug_info) and asked a formulary question. `data/` holds the trace stage by stage (`trace_stages.csv`), the deliverable as plain text (`the_trace.txt`), the four-question tools-on/tools-off grounding sweep, the per-turn token envelope, the tool registry, and this lab's registry rendered as an MCP `tools/list` response (`mcp_tools_list.json`, protocol 2026-07-28). | |
| **Every drug in this lab is invented.** The formulary is a teaching stub; no dosing information appears anywhere and nothing here is clinical guidance. | |
| **Your ~150-word takeaways go here** -- edit this README on huggingface.co. Cover: which tool-calling path your model actually used and how often the native one failed, what the tools-off answer did when it had no way to look anything up, what the prompt-based fallback cost you in tokens, and where this exact pattern shows up outside healthcare. | |