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
Make main formatter-clean so pre-commit stops rewriting every PR (#11540)
Daniel Han34094a431601
diff --git a/tests/python/test_prebuilt_cuda_wheels.py b/tests/python/test_prebuilt_cuda_wheels.pyindex f0e520ec341..ad9d94c81d8 100644--- a/tests/python/test_prebuilt_cuda_wheels.py+++ b/tests/python/test_prebuilt_cuda_wheels.py@@ -622,7 +622,7 @@ class TestWorkflow:if step.get("name", "").startswith("Smoke test"))run = smoke["run"]- assert 'Requires-Dist' in run+ assert "Requires-Dist" in runassert '"torch", "flash-attn"' in run and '"mamba-ssm"' in run and '"causal-conv1d"' in runassert "--constraint" in runassert "pip freeze | grep -E '^(torch|triton)=='" in run@@ -631,7 +631,7 @@ class TestWorkflow:lines = run.splitlines()start = next(i for i, line in enumerate(lines) if line.endswith("<<'PY'"))end = next(i for i, line in enumerate(lines) if line == "PY")- compile("\n".join(lines[start + 1:end]), "<smoke>", "exec")+ compile("\n".join(lines[start + 1 : end]), "<smoke>", "exec")def test_the_build_runs_on_the_older_ubuntu(self, workflow):"""The wheels are tagged linux_x86_64, which pip installs without a glibc check, so thediff --git a/tests/studio/test_chat_response_details_ui_contract.py b/tests/studio/test_chat_response_details_ui_contract.pyindex a14f7d4e5a2..c3b89b2e4b3 100644--- a/tests/studio/test_chat_response_details_ui_contract.py+++ b/tests/studio/test_chat_response_details_ui_contract.py@@ -624,9 +624,9 @@ def test_reasoning_clears_manual_open_on_a_new_stream():src[src.index("const [override,") :],re.S,)- assert handler and f"{writes}(open);" in handler.group(1), (- "the hand toggle must store the requested open state directly"- )+ assert handler and f"{writes}(open);" in handler.group(+ 1+ ), "the hand toggle must store the requested open state directly"# And a new round clears it. Regenerate reuses this component instance, so without this a# block opened by hand over the last answer stays pinned open over the next one.diff --git a/tests/test_omni_auto_class_resolution.py b/tests/test_omni_auto_class_resolution.pyindex e154b7f36f0..d963536097b 100644--- a/tests/test_omni_auto_class_resolution.py+++ b/tests/test_omni_auto_class_resolution.py@@ -173,11 +173,13 @@ def test_omni_reaches_the_vllm_guard_rather_than_the_language_model_path():def _tiny(cls):from transformers import LlamaConfig-return cls(LlamaConfig(- hidden_size = 4, num_hidden_layers = 1, num_attention_heads = 1,- vocab_size = 8, intermediate_size = 8,+ hidden_size = 4,+ num_hidden_layers = 1,+ num_attention_heads = 1,+ vocab_size = 8,+ intermediate_size = 8,))