Removing safetensors file

Beginners
1

Guys I keep getting this error. I deleted the previous models and yet still can’t make a twin. Help me please.

Prediction failed.

Failed to download from HuggingFace: No *.safetensors file was explicitly specified from the HuggingFace repo 1HotV/vpReplicate and more than one *.safetensors file was found. Found:

1 Like
4

Replicate related?

Here is what is going on, step by step.

1. The short version

  • Your Hugging Face model repo 1HotV/vpReplicate currently contains only a tiny .gitattributes file and no model weights at all. (Hugging Face)

  • The “twin” creation tool (which is using Hugging Face + Replicate under the hood) tries to download a *.safetensors weights file from that repo to run a prediction.

  • Because there is no .safetensors file in the repo, the loader fails and you get:

    Failed to download from HuggingFace: No *.safetensors file was explicitly specified ... Found: []

  • Deleting previous models or even deleting/recreating the repo does not fix it, because the real problem is:
    the training/export step never successfully uploaded your model weights to Hugging Face. (Hugging Face Forums)

To fix it, you need to get a real .safetensors model file into that repo (usually by re-running your training/twin creation so it properly uploads the model).

2. Breaking down the error message

Your error:

Prediction failed.
Failed to download from HuggingFace: No *.safetensors file was explicitly specified from the HuggingFace repo 1HotV/vpReplicate and more than one *.safetensors file was found. Found:

Line by line:

  1. “Prediction failed.”

    • This is just the top-level message from the service that’s trying to create or use your “twin” model. It means “we tried to run the model and it did not work.”

  2. “Failed to download from HuggingFace”

    • The backend pulls the model weights from Hugging Face before it can run any predictions. This line says that the download step (loading the model) did not succeed.

  3. **“No .safetensors file was explicitly specified … and more than one .safetensors file was found.”

    • This message comes from a helper function that scans the repo for files ending in .safetensors. In Replicate’s cog-flux weights helper, if it doesn’t find exactly one .safetensors file, it throws an error with this generic text. (GitHub)

    • It is used both when:

      • there are zero .safetensors files, or
      • there are multiple .safetensors files and the code doesn’t know which one to use.

  4. “Found: []”

    • This is the important part in your case.
    • [] means the list of .safetensors files is empty — the scanner did not find any .safetensors files in the repo at all.
    • So despite the “more than one” phrasing, the actual situation is “none found”.

So the message is a bit confusingly worded, but technically it means:

“We looked in 1HotV/vpReplicate for a .safetensors model file. We didn’t find exactly one. In fact, we found zero (empty list ). So we cannot load a model, and the prediction fails.”

3. What your Hugging Face repo actually contains

If you open your repo:

  • https://huggingface.co/1HotV/vpReplicate/tree/main

you see:

  • A single file: .gitattributes (about 1.5 kB)
  • No .safetensors files
  • No other large model weight files at all (Hugging Face)

That is what an “effectively empty” model repo looks like: it has the git metadata file but not the actual model.

On the Hugging Face forum, users who see the same error (“Failed to download from HuggingFace: No *.safetensors file…”) are in the same situation: their repository has no files because the upload from the training platform failed. (Hugging Face Forums)

4. Background: Hugging Face, safetensors, and your “twin”

4.1 Hugging Face model repos (high-level)

A Hugging Face model repository is basically an online folder that usually contains:

  • A config file (e.g. config.json)

  • One or more model weight files:

    • model.safetensors, or
    • pytorch_model.bin, or
    • Sharded weight files like model-00001-of-00002.safetensors

  • Sometimes tokenizer files, README, etc.

For your twin to run, the backend needs to download at least one of those weight files.

4.2 What is .safetensors?

  • safetensors is a file format for storing model weights safely and efficiently.
  • It’s widely used in the Hugging Face ecosystem for models, LoRAs, etc. (Hugging Face Forums)
  • Your twin service expects to find such a file in your repo (e.g., model.safetensors or something.safetensors).

If there is no .safetensors file, there is no model to load.

4.3 How your twin’s training/upload usually works

Typical pipeline:

  1. You start a “train twin” or “create AI model” process on some website/tool (often it uses Replicate as the backend).

  2. That process:

    • Trains or fine-tunes a model somewhere (e.g., on Replicate’s servers).
    • After training, uploads the resulting model weights to your Hugging Face repo (if configured to do so).

  3. When you later try to run the twin, the inference process:

    • Goes to that Hugging Face repo.
    • Looks for a .safetensors weights file.
    • Downloads it and runs the model.

If step 2 (upload to Hugging Face) fails, then:

  • The repo ends up containing no model weights.
  • The inference step still tries to download from that repo, but finds nothing.
  • You see exactly the error you pasted.

On the Hugging Face forum, a maintainer explains that this error is a download error caused by an earlier upload failure: the platform tried to upload model weights but did not succeed, leaving the repo empty. (Hugging Face Forums)

5. Why deleting “previous models” does not solve it

Deleting previous models in:

  • Your twin UI, or
  • Your local computer, or
  • Even removing and recreating the repo name

does not fix the core issue.

The core issue is:

The training/export tool that is supposed to upload your trained weights to 1HotV/vpReplicate is either misconfigured or failing silently, so no weights ever appear in the repo.

Until there is at least one valid .safetensors file in that repo, any attempt to create/use the twin will fail at the download step.

6. How to fix it in practice

Step 1 – Confirm the repo really has no weights

You already did this, but for completeness:

  1. Go to https://huggingface.co/1HotV/vpReplicate

  2. Click on the “Files and versions” tab.

  3. Check the list:

    • If you only see .gitattributes and nothing else (no large .safetensors file), then the upload never happened. (Hugging Face)

This matches your case.

Step 2 – Re-run the training / export so it actually uploads weights

You need to go back to the site or tool where you are:

  • “Creating a twin”
  • “Training your AI model”
  • Or similar (often a course platform or a front-end around Replicate).

On that platform, do the following:

  1. Check the Hugging Face connection:

    • Look for a section like “Connect Hugging Face” or “Hugging Face token”.

    • Make sure:

      • Your Hugging Face username is correct (it should be 1HotV for this repo).
      • Your Hugging Face access token is set and has permission to write to model repos.

    • If you changed your HF password, or revoked tokens recently, you might need to paste a fresh token.

  2. Check the repo name in that tool:

    • It should match exactly: 1HotV/vpReplicate (case-sensitive).
    • A typo or wrong repo (e.g. 1hotv/vpreplication) would cause the upload to go somewhere else or fail.

  3. Start a fresh training/export run:

    • Use the “Train twin”, “Create model”, or similar button again.

    • If given the choice, you can either:

      • Reuse 1HotV/vpReplicate, or
      • Pick a new name (e.g. 1HotV/vpReplicate-v2) to avoid confusion.

    • Wait until the tool claims the training/export is done.

  4. Watch Hugging Face while it runs:

    • Keep your Hugging Face model page open.
    • After the run finishes, refresh “Files and versions”.
    • A successful upload should result in at least one much larger file (likely tens or hundreds of MB) with a .safetensors extension.

If you do see such a .safetensors file appear, try your twin again. The error should disappear because the loader will now find exactly one .safetensors file.

Step 3 – If the repo is still empty after retraining

If you re-run the training/export and the Hugging Face repo still shows only .gitattributes:

  • That means the upload step is failing inside the training platform itself.

In that situation:

  1. Look for job logs:

    • Most training/hosting services have a “logs”, “run details”, or “history” page.
    • Check the log for messages about “HuggingFace”, “upload”, “token”, or “error”.

  2. Common causes:

    • Invalid or expired HF token.
    • Wrong repo name or no permission to write to that repo.
    • Quota/storage limits on the service or on Hugging Face.

  3. If the logs are unclear or you cannot find them:

    • Contact that platform’s support or ask in their help community.

    • Provide:

      • The error message you see.
      • The URL of your Hugging Face repo: https://huggingface.co/1HotV/vpReplicate.
      • A screenshot showing that the repo has no model files.

The Hugging Face forum answer for this exact error says the same: without knowing which tool is doing the upload, only that the upload failed, you need to debug or get support for that tool. (Hugging Face Forums)

Step 4 – Manual upload (only if you already have a .safetensors file)

This is optional and more advanced.

If, for some reason, you already have the trained model weights as a .safetensors file on your local machine (for example, you exported it from Replicate manually), you can:

  1. Go to your Hugging Face model: 1HotV/vpReplicate.
  2. Click “Files and versions”.
  3. Click “Upload files”.
  4. Upload your something.safetensors file.
  5. After upload completes, verify that the file appears in the list.
  6. Try your twin again.

However, in most “AI twin” or “course” setups, you never see the raw weights file directly, so manual upload is usually not available; the platform is supposed to handle it.

7. Mental model: what must be true before predictions can work

You can think of it like this:

  1. Training phase:

    • Needs to finish successfully.
    • Needs to upload a .safetensors weights file into your Hugging Face repo.

  2. Inference (prediction) phase:

    • Needs to find exactly one .safetensors file in that repo.
    • Downloads that file and uses it.

Right now, phase 2 is failing because phase 1 never produced/uploaded that file.

Once there is one real .safetensors file in 1HotV/vpReplicate, the “No *.safetensors file…” error should stop appearing.

8. Quick bullet summary

  • The error is not about “old models” on your machine.

  • Your Hugging Face repo 1HotV/vpReplicate is effectively empty: only .gitattributes, no .safetensors model file. (Hugging Face)

  • The twin backend scans the repo for *.safetensors. It finds zero, so it raises that confusing error with Found: []. (GitHub)

  • Root cause (as confirmed by Hugging Face forum discussions) is that the upload from your training/twin tool to Hugging Face failed. (Hugging Face Forums)

  • Fix: re-run training/export from the tool that makes the twin, ensuring:

    • Hugging Face token is valid and has write access.
    • Repo name matches exactly.
    • After training, the Hugging Face repo shows at least one large .safetensors file under “Files and versions”.

  • If retraining still leaves the repo empty, you need to inspect that tool’s logs or contact its support, because the failure is happening on their side before upload.

9. Useful links

  • Your model repo (shows that it currently has no weights):
    https://huggingface.co/1HotV/vpReplicate (Hugging Face)
  • Hugging Face forum: “Failed to download from HuggingFace: No *.safetensors file” – explains that the error is caused by a failed upload from an external tool (often Replicate). (Hugging Face Forums)
  • Replicate cog-flux weights helper – shows the logic that scans for .safetensors files and throws this exact type of error when it doesn’t find exactly one file. (GitHub)
  • General safetensors / HF discussion (what safetensors is and how it’s used): (Hugging Face Forums)
5

WOW thank you so much I will work on this tomorrow. You have helped me immensely. I appreciate how you went over it step by step.

1 Like