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 tests: a frozen background compile may be thawed more than once for one waiting compile (#16364)
Daniel Han1c6789c3dce7
diff --git a/studio/backend/tests/test_diffusion_bg_compile_idle.py b/studio/backend/tests/test_diffusion_bg_compile_idle.pyindex 43c60ae4361..3bb7b29c145 100644--- a/studio/backend/tests/test_diffusion_bg_compile_idle.py+++ b/studio/backend/tests/test_diffusion_bg_compile_idle.py@@ -267,7 +267,8 @@ def test_a_compile_elsewhere_thaws_a_frozen_compile_instead_of_deadlocking():worker.join(30.0)assert not worker.is_alive(), "deadlock: the render's compile waited on the frozen trace"torch.testing.assert_close(out[0], torch.full((3,), 3.0))- assert job.thaws == 1+ # At least one: a refreeze landing between the frame's lock release and the waiter's next 50 ms poll is thawed again.+ assert job.thaws >= 1# it ran on only to the end of the frame that held the lock, then froze again while the render still runs_wait_for(lambda: job._frozen, timeout = 30.0)assert job.freezes >= 2