biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix][AMD] Update torch_bindings so that scaled_fp4_quant isn't build on ROCm (#13235)

Browse Source
This commit is contained in:
Sage Moore
2025-02-14 20:30:42 -08:00
committed by GitHub
parent 0c73026844
commit c9f9d5b397
3 changed files with 12 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
csrc/torch_bindings.cpp
View File

@@ -385,6 +385,13 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
"bool silu_activation,"
"int pad_slot_id) -> ()");
ops.impl("causal_conv1d_fwd", torch::kCUDA, &causal_conv1d_fwd);
// Compute NVFP4 block quantized tensor.
ops.def(
"scaled_fp4_quant(Tensor! output, Tensor input,"
" Tensor! output_scale, Tensor input_scale) -> ()");
ops.impl("scaled_fp4_quant", torch::kCUDA, &scaled_fp4_quant);
#endif
// Quantized GEMM for GPTQ.
@@ -421,12 +428,6 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
ops.impl("dynamic_per_token_scaled_fp8_quant", torch::kCUDA,
&dynamic_per_token_scaled_fp8_quant);
// Compute NVFP4 block quantized tensor.
ops.def(
"scaled_fp4_quant(Tensor! output, Tensor input,"
" Tensor! output_scale, Tensor input_scale) -> ()");
ops.impl("scaled_fp4_quant", torch::kCUDA, &scaled_fp4_quant);
// Compute int8 quantized tensor for given scaling factor.
ops.def(
"static_scaled_int8_quant(Tensor! result, Tensor input, Tensor scale,"