5 ms·
I really like the json schemas around the tool calls. Much stricter validation than in codex. Git might be worth adding to the top level. Currently you've got
by JonChesterfield 1mo ago
I really like the json schemas around the tool calls. Much stricter validation than in codex.
Git might be worth adding to the top level. Currently you've got LSP, grep, glob nicely structured for non-mutating queries across a codebase, but git is behind bash and that means hope or sandboxing.
Thank you for uploading it. Gives a lot of insight into how the deepseek models might expect tool calls to be structured.
- youre-wrong3 1mo agoI don’t like it. Structures responses really do not work well with LLMs at all. They are one of the biggest causes of issues with tool calling right now.
- sscaryterry 1mo agoThen you're doing it wrong :)
- maleldil 1mo agoAgree with sibling. If you're getting severely deteriorated results with structured output, you're probably doing something wrong. There's been some research on the impact of structured outputs on results distribution, and there are tradeoffs, but "do not work well at all" doesn't match the experience at large.
- JonChesterfield 1mo agoYou prefer having the harness execute any markdown that looks like it might be a tool call? I had a bad time getting that to work reliably whereas a grammar in the sampler gets it right every time.
- alexgoodhart 1mo agoThis is an AI response