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
tests: make the torchcodec provenance and mirror repair tests hermetic (#12850)
Daniel Handc10bb54312f
diff --git a/tests/python/test_torchcodec_torch_compat.py b/tests/python/test_torchcodec_torch_compat.pyindex 892873506da..2448c7fbd70 100644--- a/tests/python/test_torchcodec_torch_compat.py+++ b/tests/python/test_torchcodec_torch_compat.py@@ -1069,6 +1069,13 @@ def test_a_mismatched_accelerator_build_is_named_when_the_codec_cannot_load(monkfixes = _load_import_fixes_module()_stub_torch(monkeypatch, "2.11.0+cu128")+ # Metadata is read first; report none so the stub answers, installed codec or not.+ import importlib.metadata++ def _no_metadata(name):+ raise importlib.metadata.PackageNotFoundError(name)++ monkeypatch.setattr(importlib.metadata, "version", _no_metadata)codec = types.ModuleType("torchcodec")codec.__version__ = "0.11.0" # untagged: PyPI's default buildmonkeypatch.setitem(sys.modules, "torchcodec", codec)@@ -1370,8 +1377,8 @@ def test_the_provenance_hint_reads_a_codec_it_cannot_import(monkeypatch):fixes = _load_import_fixes_module()_stub_torch(monkeypatch, "2.11.0+cu128")- # No importable torchcodec at all, which is what a failed initialisation leaves behind.- monkeypatch.delitem(sys.modules, "torchcodec", raising = False)+ # None blocks the import even with a codec installed, as a failed initialisation does.+ monkeypatch.setitem(sys.modules, "torchcodec", None)def _version(name):if name == "torchcodec":@@ -1508,6 +1515,17 @@ def test_a_query_authenticated_mirror_is_not_pinned_at_all(monkeypatch):def test_query_authenticated_mirror_repairs_decline_instead_of_falling_back(monkeypatch):"""The eight repair/flavor paths must neither mangle the token nor leave the mirror."""+ # Hidden CUDA returns before the mirror is read; index/backend pins pick another path.+ for name in (+ "CUDA_VISIBLE_DEVICES",+ "UNSLOTH_TORCH_INDEX_URL",+ "UNSLOTH_TORCH_INDEX_FAMILY",+ "UNSLOTH_TORCH_INSTALL_INDEX_URL",+ "UNSLOTH_TORCH_BACKEND",+ "UNSLOTH_ROCM_TORCH_INSTALLED",+ "UNSLOTH_FORCE_ROCM_TORCH",+ ):+ monkeypatch.delenv(name, raising = False)monkeypatch.setenv("UNSLOTH_PYTORCH_MIRROR", "https://mirror.example/whl?token=abc")mod = _reload_install_python_stack()