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 diffusion: skip the offload cost probe's pinned buffer on an integrated GPU (#15178)
Daniel Han9f499ff6478f
diff --git a/studio/backend/core/inference/diffusion_memory.py b/studio/backend/core/inference/diffusion_memory.pyindex 34e5539ee6d..ced29fa8ea0 100644--- a/studio/backend/core/inference/diffusion_memory.py+++ b/studio/backend/core/inference/diffusion_memory.py@@ -4061,7 +4061,8 @@ def _offload_cost_measurements() -> Optional[tuple[float, float]]:try:import torch- if not torch.cuda.is_available() or is_rocm_torch(torch):+ # integrated GPU: the probe's pinned buffer only takes RAM from the card; unpriced keeps the whole-module plan+ if not torch.cuda.is_available() or is_rocm_torch(torch) or not host_pinning_allowed():return Nonekey = torch.cuda.current_device()if key in _OFFLOAD_COST_PROBES: