@ivanfioravanti@jessellloyd We’ve been bumping into the same issue, inference has been slower than expected. Are you seeing tok/s > than the bf16? Testing with Qwen3.5 both 35B-A3B and 27B.
@ClementDelangue@MaziyarPanahi@cniongolo We’ve already built it - Just published in response to this post!
Cluedit - View, branch, edit your claude and codex chats and export for LLM training.
https://t.co/Iin1cMSlIi
Export a project you’ve already build as a dataset.
https://t.co/5zgwOYoBIr
@ClementDelangue@MaziyarPanahi@cniongolo We’ve already built it - Just published in response to this post!
Cluedit - View, branch, edit your claude and codex chats and export for LLM training.
https://t.co/Iin1cMSlIi
Export a project you’ve already build as a dataset.
https://t.co/5zgwOYoBIr
@ivanfioravanti https://t.co/bOUsqk5FG5 have them orchestrate with each other. Features optimistic concurrency with "expected_hash" - basically git for agent state.
@ClementDelangue I sent you a DM with a quick demo (hope that was ok). I think it's a great utility to enable anyone to output their existing agent (adding openclaw, hermes, autoresearch, etc) traces as datasets, and it outputs directly to HF datasets from the UI.
@ClementDelangue@MaziyarPanahi@cniongolo We’ve already built it - Just published in response to this post!
Cluedit - View, branch, edit your claude and codex chats and export for LLM training.
https://t.co/Iin1cMSlIi
Export a project you’ve already build as a dataset.
https://t.co/5zgwOYoBIr