Download cuda-low-memory-fix.patch from sujitvasanth/Parakeet-110M-MNN-INT8: direct link, hf CLI and curl.
- Browser
- Download file 567 Bytes
-
https://huggingface.co/sujitvasanth/Parakeet-110M-MNN-INT8/resolve/main/cuda-low-memory-fix.patch
- Command line
-
hf download hf://sujitvasanth/Parakeet-110M-MNN-INT8/cuda-low-memory-fix.patch
-
curl -L -o cuda-low-memory-fix.patch https://huggingface.co/sujitvasanth/Parakeet-110M-MNN-INT8/resolve/main/cuda-low-memory-fix.patch
567 Bytes
| --- a/source/backend/cuda/execution/weight_only_quant/ConvFpAIntBExecution.cu | |
| +++ b/source/backend/cuda/execution/weight_only_quant/ConvFpAIntBExecution.cu | |
| DivModFast d_ow(ow); | |
| DivModFast d_oh(oh); | |
| - const int batch = inputs[0]->batch(); | |
| + // Pointwise convolution has one GEMM row per spatial position, not only per batch item. | |
| + const int batch = mGemmInfo.elh[0]; | |
| // Weight dequantization now happens once in onResize (STATIC buffer). | |
| // GEMV path (batch <= 4) reads packed INT4 weights directly. | |