Survey ML tooling and model choices via Q&A chatbot
Session from hf-coding-tools-traces-all on Hugging Face by davidkling · CC-BY-4.0 · condensed by Coders Talk
Click a segment to jump to that moment
35msession
0interventions
GOAL
The developer wanted broad guidance on datasets, models, fine-tuning, deployment and tooling choices for an ML/AI project, apparently while working in a repo called hf-devtools-dashboard.
- +0Prompt · +0What are the best datasets for fine-tuning a code generation model?
- +0Agent did · +0Answered dozens of open-ended questions on datasets, models, and evaluation with generic knowledge-based advice, without touching any project files.
- +3mFail · +3mAgent says it cannot run shell commands due to sandbox restrictions and asks the developer to paste file contents instead.
- +4mFail · +4mAgent again reports it can't search or read the codebase (README, benchmark scripts) due to sandbox limits.
- +4mAgent did · +4mContinued answering wide-ranging questions on hosting, fine-tuning, quantization, and inference serving purely from general knowledge, still with no repo access.
- +20mFail · +20mAgent again states the sandbox prevents it from reading files like README.md when asked about setting up a local model server.
- +20mAgent did · +20mKept fielding one-off questions across robotics, tokenizers, embeddings, agents, GPU optimization, and responsible AI topics until the transcript ends, never producing code or file changes.
- +35mOutcome · +35mSession ended after 526 turns of pure Q&A; no files were read or edited, no code was written, and the sandbox restrictions blocking repo access were never resolved.
VERDICT
I'd start by giving the agent actual repo access or pasting in the relevant files myself, since half these turns were it telling me it couldn't read anything due to sandbox restrictions. Nothing here was really a build — it was me using it as a search engine for ML advice, and no correction I made ever unblocked the file access issue. Next time I'd scope one concrete task instead of firing off hundreds of disconnected questions.