Expand description
Ferric weight loading — a pure-Rust safetensors reader. Parses the HF safetensors container
(8-byte little-endian header length, a JSON header of name → {dtype, shape, data_offsets}, then a
flat data blob) and dequantizes F32/F16/BF16/F64 tensors to f32, ready to upload to a Ferric
Context. No Python, no C++ — this is how real Llama/SmolVLA checkpoints enter the ecosystem.
Structs§
- STensor
- One tensor decoded to f32 plus its shape.
Functions§
- safetensors
- Parse a safetensors byte buffer into name → f32 tensor.
__metadata__is skipped.