sgl-project/sglang · error · FileNotFoundError

Specified lora_weight_name '{weight_name}' not found in {loc

Error message

Specified lora_weight_name '{weight_name}' not found in {local_path}

What it means

When a LoRA is loaded from a local directory with an explicit lora_weight_name, maybe_download_lora verifies that <local_path>/<weight_name> exists as a file. If not, FileNotFoundError.

Source

Thrown at python/sglang/multimodal_gen/runtime/utils/hf_diffusers_utils.py:638

    if envs.SGLANG_USE_MODELSCOPE.get():
        allow_patterns = (
            ["*.json", weight_name, f"**/{weight_name}"]
            if weight_name is not None
            else ["*.json", "*.safetensors", "*.bin"]
        )
        local_path = maybe_download_model(
            model_name_or_path,
            local_dir,
            download,
            is_lora=True,
            allow_patterns=allow_patterns,
        )
        if os.path.isfile(local_path):
            return local_path
        if weight_name is not None:
            target = os.path.join(local_path, weight_name)
            if not os.path.isfile(target):
                raise FileNotFoundError(
                    f"Specified lora_weight_name '{weight_name}' not found in "
                    f"{local_path}"
                )
            return target
        guessed = _best_guess_weight_name(local_path, file_extension=".safetensors")
        if guessed is None and current_platform.is_rocm():
            guessed = _best_guess_weight_name(
                model_name_or_path, file_extension=".safetensors"
            )
        return os.path.join(local_path, guessed)

    resolved_weight = resolve_weight(model_name_or_path, weight_name=weight_name)
    selected_file = resolved_weight.selected_file
    if not selected_file.endswith(".safetensors"):
        raise ValueError(
            "Native diffusion LoRA loading requires a safetensors file, got "
            f"{selected_file!r}"
        )

View on GitHub (pinned to 0132848349)

Solutions

1. List the directory contents and use the exact filename: ls <local_path>
2. Fix the extension (.safetensors is required for native loading)
3. If the file is in a subfolder, point lora_path at the folder containing it or omit weight_name to trigger best-guess

Example fix

# before
load_lora_adapter("/loras/sxzl", weight_name="pytorch_lora_weights.bin")
# after
load_lora_adapter("/loras/sxzl", weight_name="pytorch_lora_weights.safetensors")

import os
target = os.path.join(local_path, weight_name)
assert os.path.isfile(target), f'{weight_name} not in {local_path}: {os.listdir(local_path)}'

try:
    load_lora_adapter(path, weight_name=w)
except FileNotFoundError:
    w = pick_safetensors(os.listdir(path))  # fallback pick
    load_lora_adapter(path, weight_name=w)

When it happens

Trigger: Calling load_lora_adapter with a local lora_path plus a lora_weight_name that does not exist in that directory (typo, wrong extension like .bin vs .safetensors, or the file lives in a subfolder).

Common situations: Typos in weight names; copying only some files from a LoRA repo; expecting .bin when only .safetensors was published.

Related errors

• metal shader source not found: {metal_src}
• LoRA batch_info must provide max_len or seg_lens.
• LoRA batch_info must provide max_len or seg_lens.
• lora_nickname cannot be empty
• Failed to set LoRA adapter: {str(e)}

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/aefb5edfd2bcdac5. Report an issue: GitHub.