`ptwm_core`

Pure-Rust core for lossless neural-network weight compression.

rsmod

ptwm_core

mod ptwm_core

Pure-Rust core for lossless neural-network weight compression.

No Python dependency lives here; the sibling `ptwm-py` crate carries the PyO3 bindings. The modules below implement the `.ptwm` container format (header, prelude, index, tensor / plane records, compressor, container reader) and the codec plus preprocessing utilities the container and the Python CLI share. # API stability The stable, supported entry point to PTWM is the **Python package** (`pip install ptwm`). This crate's API is intended for the sibling [`ptwm-py`] bindings and may change in any minor release. External Rust consumers should pin an exact version. [`ptwm-py`]: https://crates.io/crates/ptwm-py

rsfn

ptwm_core::buffer_sizes_for_chunk

crates/ptwm-core/src/lib.rs:148

pub fn buffer_sizes_for_chunk(chunk_len: usize, num_buf: usize, bytes_mode: i32) -> Vec<usize>

Compute per-buffer sizes for a chunk of `chunk_len` bytes.

rsfn

ptwm_core::combine_chunk

crates/ptwm-core/src/lib.rs:118

pub fn combine_chunk(planes: &[Vec<u8>], out: &mut [u8], num_buf: usize, bits_mode: i32, bytes_mode: i32)

Inverse of [`split_chunk`]. Reassembles `planes` into the output buffer.

`out` must be sized to the original uncompressed chunk length (before splitting). `planes.len()` must equal `num_buf`.

rsfn

ptwm_core::split_chunk

crates/ptwm-core/src/lib.rs:99

pub fn split_chunk(data: &[u8], num_buf: usize, bits_mode: i32, bytes_mode: i32) -> Vec<Vec<u8>>

Split a single chunk's data into `num_buf` buffers based on dtype.

rsfn

ptwm_core::validate_modes

crates/ptwm-core/src/lib.rs:63

pub fn validate_modes(num_buf: usize, bits_mode: i32, bytes_mode: i32) -> Result<(), PtwmCoreError>

Validate `bits_mode` / `bytes_mode` for the given `num_buf`.

rsfn

ptwm_core::validate_num_buf

crates/ptwm-core/src/lib.rs:55

pub fn validate_num_buf(num_buf: u32) -> Result<usize, PtwmCoreError>

Validate `num_buf` and return it as a `usize`.