Issues · unslothai/unsloth · GitHub

Labels Milestones New issue

[Bug] Gemma 4 31B grad_norm spikes to ~11,000 on a normal-looking step during LoRA fine-tuning

#5207

· pritmish opened

on Apr 28, 2026

[Bug] POST request to `/load` in Unsloth Studio does not respond until the model is fully loaded, causing proxies like Cloudflare to time out

#5206

· markasoftware-tc opened

on Apr 27, 2026

[Bug] DPOTrainer + Gemma 4 vision: dataset.map hangs at "Tokenizing 0/52" step (process_row works in isolation)

#5196

· shahfazal opened

on Apr 27, 2026

[Feature request] [Studio] There is no way to import and export the dataset

#5193

· VeryEvilHumna opened

on Apr 26, 2026

[Feature] - "merged float32" - Float32 precision safetensors, with low VRAM/RAM requirements.

feature request

#5186

· David-AU-github opened

on Apr 26, 2026

importing unsloth before trl on multi image grpo training gives mismatch error

currently fixing

#5183

· jaaabir opened

on Apr 25, 2026

[Feature] Install / download location

feature request

#5182

· JamesHutchison opened

on Apr 25, 2026

[Bug] Unsloth detected signs of an AMD ROCm GPU, but your current PyTorch build has no usable HIP accelerator

#5180

· guinmoon opened

on Apr 25, 2026

[Feature] SSL Secure Connection

feature request

#5177

· ctsstc opened

on Apr 24, 2026

[Bug] RuntimeError: shape mismatch in get_rope_index during GRPO training with Qwen3-VL(8B)

#5176

· Saad1926Q opened

on Apr 24, 2026

[Bug] Latest Unsloth Studio UI on phone

#5166

· eroji opened

on Apr 24, 2026

[Bug] Unable to resume training from existing checkpoint [QWEN3.6]

#5165

· imezx opened

on Apr 24, 2026