Nemotron-Nano-30B: What settings are you getting good results with?

A Reddit user is seeking advice on the optimal settings for running the Nemotron-Nano-30B language model for their use case of agentic coding with Qwen-Code-CLI.

💡

Why it matters

Optimizing language model settings is crucial for achieving reliable and consistent results, especially for mission-critical applications like agentic coding.

Key Points

  • 1The user is currently using temperature=0.6, top_p=0.95, and top_k 20 settings
  • 2The model performs well initially, but starts exhibiting issues like infinite retry loops and
  • 3 behavior after about 50k tokens
  • 4The user is looking for recommendations on better settings to improve the model's performance for their agentic coding use case

Details

The Nemotron-Nano-30B is a large language model that the user is attempting to use for agentic coding with the Qwen-Code-CLI tool. They have tried a set of common settings, including temperature, top_p, and top_k, but are encountering issues with the model's behavior after generating around 50,000 tokens. The user is seeking advice from the community on alternative settings or techniques that could help improve the model's performance and reliability for their specific use case.

Like
Save
Read original
Cached
Comments
?

No comments yet

Be the first to comment

AI Curator - Daily AI News Curation

AI Curator

Your AI news assistant

Ask me anything about AI

I can help you understand AI news, trends, and technologies