[KERNEL] Sampler. CUDA kernel for applying repetition penalty (#18437)
This commit is contained in:
@@ -170,6 +170,13 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
|
||||
"float epsilon) -> ()");
|
||||
ops.impl("fused_add_rms_norm", torch::kCUDA, &fused_add_rms_norm);
|
||||
|
||||
// Apply repetition penalties to logits in-place
|
||||
ops.def(
|
||||
"apply_repetition_penalties_(Tensor! logits, Tensor prompt_mask, "
|
||||
"Tensor output_mask, Tensor repetition_penalties) -> ()");
|
||||
ops.impl("apply_repetition_penalties_", torch::kCUDA,
|
||||
&apply_repetition_penalties_);
|
||||
|
||||
// Layernorm-quant
|
||||
// Apply Root Mean Square (RMS) Normalization to the input tensor.
|
||||
ops.def(
|
||||
|
||||
Reference in New Issue
Block a user