Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
Studio ROCm TunableOp: label the results file with the HIP build string, not a raw read (#17325)
Daniel Han1ae69bfe6a8a
diff --git a/studio/backend/core/inference/diffusion_rocm_tunableop.py b/studio/backend/core/inference/diffusion_rocm_tunableop.pyindex 5b4c2a1a75a..c454a8c26db 100644--- a/studio/backend/core/inference/diffusion_rocm_tunableop.py+++ b/studio/backend/core/inference/diffusion_rocm_tunableop.py@@ -124,9 +124,9 @@ def results_file(arch: str) -> Path:from .diffusion_compile_cache import cache_root- tag = "_".join(- re.sub(r"[^A-Za-z0-9.]+", "-", str(x)) for x in (arch, torch.__version__, torch.version.hip)- )+ # The HIP build string only labels the file (an f-string, as the ROCm gate guard expects); None on CUDA.+ hip = f"{getattr(torch.version, 'hip', None)}"+ tag = "_".join(re.sub(r"[^A-Za-z0-9.]+", "-", str(x)) for x in (arch, torch.__version__, hip))return cache_root() / "tunableop" / f"{tag}.csv"