List tensors, dtypes and shapes in a checkpoint.
Drop your model file here
or pick it from your device
Only the file's header bytes are read in your browser — the full checkpoint never leaves your device.
How to use Safetensors Inspector
Add a safetensors file. Drop a .safetensors file onto the uploader or pick it from your device. Only one file is accepted at a time.
Wait for the header to be read. The tool reads the start of the file and shows Reading header while it parses the JSON header.
Check the tensor count. The safetensors header card shows how many tensors the file contains.
Read the metadata. If the file has a __metadata__ block, its key-value pairs appear in a Metadata table.
Browse the tensors. Scroll the Tensors table to see each tensor's name, shape and dtype.
Limits and privacy
- Only the beginning of the file is read, never the tensor data. It starts with a 4 MB read and doubles the window up to a 64 MB ceiling if the header is larger.
- A safetensors header declaring a length over 64 MB is rejected as implausible.
- The upload limit is 50 GB per file, and only one file at a time.
- The file must have a .safetensors or .gguf extension. ONNX, .bin and .pt files are not supported by this inspector.
- The table shows name, shape and dtype. Tensor byte sizes are parsed but not displayed.
- Sharded checkpoints are inspected one shard file at a time.
- Processing happens in your browser. The file is not uploaded.
Frequently asked questions
How do I check tensor shapes in a safetensors file without loading it?
Drop the file here. The tool reads only the JSON header at the start of the file and lists every tensor's name, shape and dtype.
Can I see the dtype of each weight?
Yes. The Dtype column shows the value stored in the header for each tensor, such as F16 or BF16.
Does this work on multi-gigabyte files?
Yes, up to the 50 GB upload limit, because only the header is read and not the weights.
Is the file sent to a server?
No. Parsing runs locally in your browser.