google-research/timesfm · error · FileNotFoundError

{cls.WEIGHTS_FILENAME} not found in directory {model_id}

Error message

{cls.WEIGHTS_FILENAME} not found in directory {model_id}

What it means

_from_pretrained is the from_pretrained hook: when model_id points to a local directory it requires model.safetensors (cls.WEIGHTS_FILENAME) inside it; otherwise it downloads from Hugging Face via hf_hub_download. If model_id is a directory lacking the weights file, a FileNotFoundError is raised before any model is built.

Source

Thrown at src/timesfm/timesfm_2p5/timesfm_2p5_torch.py:341

    try:
      hf_hub_download(
          repo_id=model_id,
          filename=cls.CONFIG_FILENAME,
          revision=revision,
          cache_dir=cache_dir,
          force_download=force_download,
          local_files_only=local_files_only,
          token=token,
      )
    except Exception:
      pass
    
    model_file_path = ""
    if os.path.isdir(model_id):
      logging.info("Loading checkpoint from local directory: %s", model_id)
      model_file_path = os.path.join(model_id, cls.WEIGHTS_FILENAME)
      if not os.path.exists(model_file_path):
        raise FileNotFoundError(
          f"{cls.WEIGHTS_FILENAME} not found in directory {model_id}"
        )
    else:
      logging.info("Downloading checkpoint from Hugging Face repo %s", model_id)
      model_file_path = hf_hub_download(
        repo_id=model_id,
        filename=cls.WEIGHTS_FILENAME,
        revision=revision,
        cache_dir=cache_dir,
        force_download=force_download,
        token=token,
        local_files_only=local_files_only,
      )

    # Create an instance of the model wrapper class.
    instance = cls(config=config, **model_kwargs)

    logging.info("Loading checkpoint from: %s", model_file_path)

View on GitHub (pinned to 331c6d33cb)

Solutions

1. Check the directory contains model.safetensors; if not, re-download it from the Hugging Face repo (with git-lfs enabled or huggingface-cli download).
2. Install/initialize git-lfs and pull real weights: git lfs install && git lfs pull in the cloned repo.
3. Pass the correct HF repo id (e.g. google/timesfm-2.5-200m-pytorch) or the direct .safetensors file path instead of the incomplete directory.
4. Verify the download completed (file size matches the repo listing) before loading.

Example fix

// before
model = TimesFM_2p5_200M_torch.from_pretrained("./saved_model")  # dir lacks model.safetensors
// after
model = TimesFM_2p5_200M_torch.from_pretrained("google/timesfm-2.5-200m-pytorch")
# or ensure ./saved_model/model.safetensors exists before calling

import os
if os.path.isdir(model_id) and not os.path.exists(os.path.join(model_id, "model.safetensors")):
    model_id = "google/timesfm-2.5-200m-pytorch"  # fall back to HF repo id

try:
    model = TimesFM_2p5_200M_torch.from_pretrained(model_id)
except FileNotFoundError:
    logging.warning("Local dir %s lacks model.safetensors; using HF repo", model_id)
    model = TimesFM_2p5_200M_torch.from_pretrained("google/timesfm-2.5-200m-pytorch")

When it happens

Trigger: Calling TimesFM_2p5_200M_torch.from_pretrained(local_dir) where local_dir is a directory without model.safetensors. A non-directory model_id is instead treated as a Hugging Face repo id and downloaded.

Common situations: A local HF snapshot cache missing the LFS weights, cloning the model repo without git-lfs, pointing from_pretrained at a directory containing config/tokenizer files but not weights, or an interrupted copy of a checkpoint directory.

Related errors

• {self.WEIGHTS_FILENAME} not found in directory {path}

AI-assisted analysis of google-research/timesfm@331c6d33cb (2026-08-29). Data as JSON: /api/errors/9d1638a7ba8ff149. Report an issue: GitHub.