earl-of-embedding comments

Results 5 comments of


                                            earl-of-embedding

Codex with gpt-oss:120b model appears to forget the task

Theory: Could it be too short context in ollama from codex cli? Why do I think that? If I run gpt-oss:20b, long documents also de-rail codex (ubuntu linux machine). "ollama...

Codex with gpt-oss:120b model appears to forget the task

I am still scratching my head. https://github.com/openai/codex/blob/main/codex-rs/core/src/openai_model_info.rs clearly states that 128k should be used for context: ```rust pub(crate) fn get_model_info(model_family: &ModelFamily) -> Option { let slug = model_family.slug.as_str(); match slug...

Codex with gpt-oss:120b model appears to forget the task

You are right. In the meantime, I tried to correct this behavior in my local fork of codex-rs. With a lot of tinkering I got "ollama ps" sometimes to show...

Codex with gpt-oss:120b model appears to forget the task

Curious. When you do "ollama ps" during codex operation, is the model always with 32k or does it sometimes 'switch back' to 8K?

Codex with gpt-oss:120b model appears to forget the task

Just for other readers of this thread: I had success with running llamacpp with full 128k of both the 20B and the 120B model of GPT-OSS-x20B (quad 3090 system here)....