LoupGarou

/

deepseek-coder-6.7b-instruct-pythagora-gguf

like 0

These 6.7b models work best when used with the accompanying proxy script from GitHub: Pythagora-LLM-Proxy. I recommend you check there and use the latest model to ensure proper error handling and formatting.

Here's current list of error handling to address known issues:

1. Re-query missing plans: Occasionally, the model may not produce a plan and instead pass "plan": [] which causes Pythagora to report 100% complete.

2. Re-query missing tasks: Under certain circumstances, the model may not generate tasks when necessary and instead pass "tasks": [] which may cause issues with application development.

3. Re-query repetitive questions: During the initial planning stage, when the project description is short, the model may repeat the same questions and get stuck in a loop.