Dogacel

/

Qwen3-8B-DSpark

like 0

Thank you,
Any performance metrics to be shared?
Can this be used with llama.cpp after the conversion to gguf?

Thanks again.

This is an early checkpoint, I did not run eval code but the training acceptance length was around ~2 (excluding bonus token).

llama.cpp is always slow to catchup, I don't think it will get this model for a long time.

PR on llama.cpp: https://github.com/ggml-org/llama.cpp/pull/25173