A lightweight Python GUI application for inspecting .tbm model container files produced by tensorbit-core. Requires no external dependencies — uses only the Python standard library (tkinter).
┌─────────────────────┐
│ Tensor 0 (.tb) │ 4096-byte TBHeader + FP32 weights + mask bytes
├─────────────────────┤
│ Tensor 1 (.tb) │
├─────────────────────┤
│ ... │
├─────────────────────┤
│ Tensor K (.tb) │
├─────────────────────┤
│ JSON Index │ UTF-8 JSON: architecture, tensor names, shapes,
│ │ byte offsets, N:M params, dtypes, weight/mask counts
├─────────────────────┤
│ 4-byte LE uint32 │ Byte length of the JSON index (little-endian)
└─────────────────────┘
The JSON index enables random-access reading of individual tensors without loading the entire file into memory.
python tbm_viewer.pyClick Open .tbm file... or press Ctrl+O to choose a .tbm file.
The left panel lists all tensors with a search/filter bar. Select a tensor to view:
- Tensor Details — metadata (name, shape, dtype, N:M sparsity, byte offsets, data sizes)
- Weight Preview — first N floating-point values (configurable, max 1024)
- Mask Preview — first 256 mask bytes in hex dump format
Apache 2.0. See LICENSE.