AMD AI Reinforcement Learning Hackathon with Unsloth

​​Learn hands-on techniques for ​Reinforcement Learning for AI models with Unsloth from Daniel Han, the creator of Unsloth.

You can view Unsloth's GitHub repo here: https://github.com/unslothai/unsloth

Here is the link to our AMD fine-tuning notebooks:

notebooks/nb/gpt_oss_(20B)_Reinforcement_Learning_2048_Game_BF16.ipynb at main · unslothai/notebooksGitHub

https://github.com/unslothai/notebooks/blob/main/nb/gpt_oss_(20B)_Reinforcement_Learning_2048_Game_BF16.ipynb

wget 'https://raw.githubusercontent.com/unslothai/notebooks/refs/heads/main/nb/gpt_oss_(20B)_Reinforcement_Learning_2048_Game_BF16.ipynb'

If wanting to upgrade Unsloth / Unsloth Zoo:

uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.0 --upgrade --force-reinstall

pip uninstall unsloth unsloth_zoo -y && \
    pip install git+https://github.com/unslothai/unsloth-zoo git+https://github.com/unslothai/unsloth --no-deps --force-reinstall --no-cache-dir

For bitsandbytes:

pip install "unsloth[amd] @ git+https://github.com/unslothai/unsloth"

If you see:

error: Failed to install: bitsandbytes-1.33.7rc0-py3-none-manylinux_2_24_x86_64.whl (bitsandbytes==1.33.7rc0 (from https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-manylinux_2_24_x86_64.whl)) Caused by: Wheel version does not match filename (0.49.2.dev0 != 1.33.7rc0), which indicates a malformed wheel. If this is intentional, set UV_SKIP_WHEEL_FILENAME_CHECK=1.

Do NOT use UV_SKIP_WHEEL_FILENAME_CHECK, instead ONLY use pip install "unsloth[amd] @ git+https://github.com/unslothai/unsloth" (NOT uv) since uv destroys bitsandbytes. Maybe add a check to the PRs if possible to catch these.

For AMD installation instructions, you can view our guide here:

AMD