GitHub · Where software is built

8000 GitHub · Where software is built

Labels Milestones New issue

[Bug] Qwen3.5 vision training , error "only for image model"

#4202

· coolbei opened

on Mar 11, 2026

requires device with capability <= (9, 0) but your GPU has capability (12, 0) (too new)

#4200

· Arifursdev opened

on Mar 11, 2026

[Feature] add aarch64 container images

feature request

#4198

· zita-ch opened

on Mar 11, 2026

[Bug] Smart gradient offloading crashes with DeepSpeed ZeRO Stage 1/2 (grad_reduc is None)

#4195

· janbernloehr opened

on Mar 10, 2026

[Bug] Starcoder2-7b crashes with flex_attention ValueError on non-zero attention_dropout since v2026.2.1

#4194

· janbernloehr opened

on Mar 10, 2026

[Bug] Unsloth does not trained MoE Layer on GPT OSS

#4193

· angkywilliam opened

on Mar 9, 2026

TRL Version Bump

#4190

· d4nieldev opened

on Mar 9, 2026

[Bug] Please fill in your issue title here.

insufficient information

#4189

· Samarth26 opened

on Mar 9, 2026

[Bug] Extremely high CPU/VRAM usage and slow training with Qwen3.5

#4188

· XueNianOfficial opened

on Mar 9, 2026

GRPO off-by-one shape mismatch in UnslothGRPOTrainer masked_batch_mean (25 vs 24)

#4181

· Atharva2099 opened

on Mar 8, 2026

Is Vision DPO Training supported?

#4174

· HamzaSulaiman1996 opened

on Mar 6, 2026

[Bug] Unable to finetune older models like `SmolLM2` and `Qwen2` with unsloth

#4172

· ParagEkbote opened

on Mar 6, 2026

0