Yeah the small models really fail compared to SOTA paid for ones…
Not even the 70b or 32b coding ones are passing the test. Not going to replace windsurf with it just yet… sad
I'm running this qwen 7b param model for a local coding-ai goose agent, seems to be working ok, at least for boilerplate stuff so far.
https://ollama.com/hhao/qwen2.5-coder-tools
Yeah the small models really fail compared to SOTA paid for ones…
Not even the 70b or 32b coding ones are passing the test. Not going to replace windsurf with it just yet… sad
No replies yet.