SOTA Open Source TTS
Reuse cache for different generation size
Lengyue0dafcf373de8
diff --git a/tools/llama/generate.py b/tools/llama/generate.pyindex e7b5cce..c84a69b 100644--- a/tools/llama/generate.py+++ b/tools/llama/generate.py@@ -163,7 +163,7 @@ def decode_n_tokens(**sampling_kwargs,):previous_tokens = torch.zeros(- (model.config.num_codebooks + 1, num_new_tokens),+ (model.config.num_codebooks + 1, model.config.max_seq_len),dtype=torch.int,device=cur_token.device,)