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
Risky loader baseline: record the FSDP2 export's store handoff (#15379)
Daniel Han208d46be3f95
diff --git a/scripts/risky_loaders_baseline.json b/scripts/risky_loaders_baseline.jsonindex d1ff8670aca..f70a419c3cb 100644--- a/scripts/risky_loaders_baseline.json+++ b/scripts/risky_loaders_baseline.json@@ -626,6 +626,14 @@"count": 1,"reason": "Falls back to the auto class unless the caller passed trust_remote_code; revision pinned to the config commit hash."},+ {+ "file": "unsloth/models/fsdp2_save.py",+ "rule": "unsafe-deserialize",+ "sink": "pickle.loads",+ "digest": "bf2d3c412d57fd50",+ "count": 1,+ "reason": "_share_from_writer: unpickles only what the export's writer rank stored under an unsloth_fsdp2_export_* key in the job's own default c10d store; a party able to write that store already controls the job's collectives."+ },{"file": "unsloth/models/loader.py","rule": "dynamic-import",