Note: Please do not remove the questions. Answer beside them.
- Did you update? pip install --upgrade unsloth unsloth_zoo — Yes, fresh install via curl -fsSL https://unsloth.ai/install.sh | sh today (2026.7.4)
- Colab or Kaggle or local / cloud — Local
- Number GPUs used, use nvidia-smi — 1 GPU (AMD, not NVIDIA — nvidia-smi not applicable). rocminfo shows: AMD Ryzen AI MAX+ 395 w/ Radeon 8060S, gfx1100, ROCm HIP 6.3.42134
- Which notebook? Please link! — Not using a notebook. Running Unsloth Studio via unsloth studio -p 8888 on Linux Mint desktop.
- Which Unsloth version, TRL version, transformers version, PyTorch version? — Unsloth: 2026.7.4 | Transformers: 5.10.2 | PyTorch: 2.9.1+rocm6.3 | TRL: installed via Unsloth Studio installer
- Which trainer? SFTTrainer, GRPOTrainer etc — Not reached training yet. Studio crashes on startup before any trainer is selected, in the RAG embeddings warmup thread (studio/backend/core/rag/embeddings.py → SentenceTransformer.init → .to(device) → segfault)
Minimal reproduction: Unsloth Studio RAG embeddings warmup segfault on AMD ROCm
No HF token needed — crash happens before any model is loaded from HF
Environment:
OS: Linux Mint
GPU: AMD Radeon 8060S (gfx1100), integrated/unified memory, 128GB RAM
ROCm: 6.3.0
PyTorch: 2.9.1+rocm6.3
Unsloth: 2026.7.4
To reproduce:
2. Create /opt/rocm symlink if not present: sudo ln -s /opt/rocm-6.3.0 /opt/rocm
3. Run: unsloth studio -p 8888
Studio starts, server comes up, then segfaults in background thread
The crash can also be reproduced directly:
from sentence_transformers import SentenceTransformer
import torch
model = SentenceTransformer("all-MiniLM-L6-v2")
model.to("cuda") # .to(device) on AMD ROCm GPU causes segfault
Expected: model moves to GPU
Actual: Segmentation fault (core dumped)
🦥 You can also ask via our Reddit page: https://reddit.com/r/unsloth/
Note: Please do not remove the questions. Answer beside them.
Minimal reproduction: Unsloth Studio RAG embeddings warmup segfault on AMD ROCm
No HF token needed — crash happens before any model is loaded from HF
Environment:
OS: Linux Mint
GPU: AMD Radeon 8060S (gfx1100), integrated/unified memory, 128GB RAM
ROCm: 6.3.0
PyTorch: 2.9.1+rocm6.3
Unsloth: 2026.7.4
To reproduce:
1. Install Unsloth Studio: curl -fsSL https://unsloth.ai/install.sh | sh
2. Create /opt/rocm symlink if not present: sudo ln -s /opt/rocm-6.3.0 /opt/rocm
3. Run: unsloth studio -p 8888
Studio starts, server comes up, then segfaults in background thread
The crash can also be reproduced directly:
from sentence_transformers import SentenceTransformer
import torch
model = SentenceTransformer("all-MiniLM-L6-v2")
model.to("cuda") # .to(device) on AMD ROCm GPU causes segfault
Expected: model moves to GPU
Actual: Segmentation fault (core dumped)