David-AU-github

Same issue too ; Qwen 3 as well. Can load in 4 bit or 16 bit - no issues. Also updated unsloth too ; no change in issues. Attempted to...

See note on this doc (I had same issue: low training on Qwen3 MOE ; except if you add lm_head): https://docs.unsloth.ai/models/qwen3-how-to-run-and-fine-tune Qwen3 MOE models fine-tuning Fine-tuning support includes MOE models:...

> > See note on this doc (I had same issue: low training on Qwen3 MOE ; except if you add lm_head): > > https://docs.unsloth.ai/models/qwen3-how-to-run-and-fine-tune > > Qwen3 MOE models...