{"record":{"id":"26b757a3bfaa896a","repo":"sgl-project/sglang","slug":"minimax-h3-adaln-online-rebuilds-adaln-outputs-f","errorCode":null,"errorMessage":"--minimax-h3-adaln-online rebuilds AdaLN outputs from the safetensors checkpoint and cannot read a GGUF transformer.","messagePattern":"--minimax-h3-adaln-online rebuilds AdaLN outputs from the safetensors checkpoint and cannot read a GGUF transformer\\.","errorType":"validation","errorClass":"ValueError","httpStatus":null,"severity":"error","filePath":"python/sglang/multimodal_gen/runtime/loader/transformer_load_utils.py","lineNumber":571,"sourceCode":"        else server_args.use_fsdp_inference\n    )\n    if uses_fsdp:\n        raise ValueError(\n            \"GGUF diffusion checkpoints are incompatible with FSDP inference. \"\n            \"Run without --use-fsdp-inference, or keep this component offloaded \"\n            \"so FSDP does not manage it.\"\n        )\n    if server_args.lora_path is not None:\n        raise ValueError(\n            \"LoRA is not supported on a GGUF transformer: an adapter cannot be \"\n            \"merged into packed GGML blocks. Use the unquantized checkpoint to \"\n            \"serve LoRA.\"\n        )\n    # H3's AdaLN paths read the transformer's safetensors directly -- the cache\n    # builder needs unquantized weights, and the online rebuild is handed the\n    # safetensors file list, which is empty for a GGUF load.\n    if server_args.minimax_h3_adaln_online:\n        raise ValueError(\n            \"--minimax-h3-adaln-online rebuilds AdaLN outputs from the \"\n            \"safetensors checkpoint and cannot read a GGUF transformer.\"\n        )\n    if server_args.minimax_h3_adaln_cache_path is not None:\n        raise ValueError(\n            \"--minimax-h3-adaln-cache-path requires the unquantized \"\n            \"transformer and cannot be combined with a GGUF checkpoint.\"\n        )\n\n\ndef resolve_transformer_gguf_to_load(\n    server_args: ServerArgs, component_name: str | None = None\n) -> Optional[str]:\n    \"\"\"Resolve ``--transformer-weights-path`` to a local ``.gguf``, if it is one.\n\n    Returns ``None`` when the override is absent or is not GGUF, so the caller\n    falls through to the safetensors path.\n    \"\"\"","sourceCodeStart":553,"sourceCodeEnd":589,"githubUrl":"https://github.com/sgl-project/sglang/blob/0132848349585cfe6aae51c4941cbae872505f8a/python/sglang/multimodal_gen/runtime/loader/transformer_load_utils.py#L553-L589","documentation":"The --minimax-h3-adaln-online path rebuilds AdaLN outputs by reading the transformer's safetensors directly; a GGUF load has no safetensors file list, so the rebuild cannot run and is rejected.","triggerScenarios":"server_args.minimax_h3_adaln_online=True combined with a GGUF transformer checkpoint.","commonSituations":"Enabling the H3 AdaLN online rebuild for faster startup and then switching the transformer to a GGUF quantized export.","solutions":["Disable --minimax-h3-adaln-online when using GGUF","Use an unquantized safetensors transformer if the online AdaLN rebuild is required"],"exampleFix":"# before\n--minimax-h3-adaln-online --transformer-weights-path model.gguf\n# after\n--transformer-weights-path model.gguf","handlingStrategy":"validation","validationCode":"if server_args.minimax_h3_adaln_online and str(server_args.transformer_weights_path or '').endswith('.gguf'):\n    raise SystemExit('--minimax-h3-adaln-online needs a safetensors transformer')","typeGuard":null,"tryCatchPattern":null,"preventionTips":["The H3 AdaLN rebuild/cache flags assume safetensors; disable them for GGUF"],"tags":["gguf","minimax-h3","adaln","config-conflict"],"backgroundTag":"incompatible-launch-flags","analyzedSha":"0132848349585cfe6aae51c4941cbae872505f8a","analyzedAt":"2026-08-28T05:10:05.995Z","schemaVersion":2},"datasetVersion":"2026-08-28T06:17:29.519Z"}