mirror of https://github.com/tracel-ai/burn.git
eb899db16c
* Add q_* ops to match float ops * Refactor q_* ops w/ dequant_op_quant macro * Comparison ops are already implemented by default to compare dequantized values * Add default arg min/max implementation and fix tch implementation * Avoid division by zero scale * Add default q_gather implementation (tch does not support on quantized tensor) * Add warning instead for tch quantize_dynamic * Call chunk backend implementation * Add QFloat check for q_ ops * Add tch q_min/max_dim_with_indices * Add q_ ops tests * Clippy fix * Remove dead code/comments * Fix quantization tests precision * Set higher tolerance for ndarray backend * Remove comment |
||
---|---|---|
.. | ||
src | ||
Cargo.toml | ||
LICENSE-APACHE | ||
LICENSE-MIT | ||
README.md | ||
build.rs |
README.md
Burn NdArray
Burn ndarray backend
Feature Flags
This crate can be used without the standard library (#![no_std]
) with alloc
by disabling the
default std
feature.
The following flags support various BLAS options:
blas-accelerate
- Accelerate framework (macOS only)blas-netlib
- Netlibblas-openblas
- OpenBLAS static linkedblas-openblas-system
- OpenBLAS from the system
Note: under the no_std
mode, the seed is fixed if the seed is not
initialized by by Backend::seed
method.
Platform Support
Option | CPU | GPU | Linux | MacOS | Windows | Android | iOS | WASM |
---|---|---|---|---|---|---|---|---|
Pure Rust | Yes | No | Yes | Yes | Yes | Yes | Yes | Yes |
Accelerate | Yes | No | No | Yes | No | No | Yes | No |
Netlib | Yes | No | Yes | Yes | Yes | No | No | No |
Openblas | Yes | No | Yes | Yes | Yes | Yes | Yes | No |