From f21c4764ecb04cd5108252802c568f452fe2d576 Mon Sep 17 00:00:00 2001 From: "lyuxiang.lx" Date: Thu, 11 Dec 2025 06:04:36 +0000 Subject: [PATCH] fix bug --- cosyvoice/llm/llm.py | 2 +- cosyvoice/utils/common.py | 52 +++++++++++++++++++-------------------- 2 files changed, 27 insertions(+), 27 deletions(-) diff --git a/cosyvoice/llm/llm.py b/cosyvoice/llm/llm.py index a343ce4..6b3a7b0 100644 --- a/cosyvoice/llm/llm.py +++ b/cosyvoice/llm/llm.py @@ -604,7 +604,7 @@ class Qwen2LM(TransformerLM): masks=torch.tril(torch.ones((1, seq_len, seq_len), device=lm_input.device)).to(torch.bool), cache=cache) logp = self.llm_decoder(y_pred[:, -1]).log_softmax(dim=-1) - top_ids = self.sampling_ids(logp.squeeze(dim=0), out_tokens, sampling, ignore_eos=False).item() + top_ids = self.sampling_ids(logp.squeeze(dim=0), out_tokens, sampling, ignore_eos=False) out_tokens.append(top_ids) if top_ids >= self.speech_token_size: if top_ids == self.eos_token: diff --git a/cosyvoice/utils/common.py b/cosyvoice/utils/common.py index 8afe308..5d307ae 100644 --- a/cosyvoice/utils/common.py +++ b/cosyvoice/utils/common.py @@ -25,32 +25,32 @@ import torch IGNORE_ID = -1 -instruct_list = ["You are a helpful assistant. 请用广东话表达。", - "You are a helpful assistant. 请用东北话表达。", - "You are a helpful assistant. 请用甘肃话表达。", - "You are a helpful assistant. 请用贵州话表达。", - "You are a helpful assistant. 请用河南话表达。", - "You are a helpful assistant. 请用湖北话表达。", - "You are a helpful assistant. 请用湖南话表达。", - "You are a helpful assistant. 请用江西话表达。", - "You are a helpful assistant. 请用闽南话表达。", - "You are a helpful assistant. 请用宁夏话表达。", - "You are a helpful assistant. 请用山西话表达。", - "You are a helpful assistant. 请用陕西话表达。", - "You are a helpful assistant. 请用山东话表达。", - "You are a helpful assistant. 请用上海话表达。", - "You are a helpful assistant. 请用四川话表达。", - "You are a helpful assistant. 请用天津话表达。", - "You are a helpful assistant. 请用云南话表达。", - "You are a helpful assistant. Please say a sentence as loudly as possible.", - "You are a helpful assistant. Please say a sentence in a very soft voice.", - "You are a helpful assistant. 请用尽可能慢地语速说一句话。", - "You are a helpful assistant. 请用尽可能快地语速说一句话。", - "You are a helpful assistant. 请非常开心地说一句话。", - "You are a helpful assistant. 请非常伤心地说一句话。", - "You are a helpful assistant. 请非常生气地说一句话。", - "You are a helpful assistant. 我想体验一下小猪佩奇风格，可以吗？", - "You are a helpful assistant. 你可以尝试用机器人的方式解答吗？"] +instruct_list = ["You are a helpful assistant. 请用广东话表达。<|endofprompt|>", + "You are a helpful assistant. 请用东北话表达。<|endofprompt|>", + "You are a helpful assistant. 请用甘肃话表达。<|endofprompt|>", + "You are a helpful assistant. 请用贵州话表达。<|endofprompt|>", + "You are a helpful assistant. 请用河南话表达。<|endofprompt|>", + "You are a helpful assistant. 请用湖北话表达。<|endofprompt|>", + "You are a helpful assistant. 请用湖南话表达。<|endofprompt|>", + "You are a helpful assistant. 请用江西话表达。<|endofprompt|>", + "You are a helpful assistant. 请用闽南话表达。<|endofprompt|>", + "You are a helpful assistant. 请用宁夏话表达。<|endofprompt|>", + "You are a helpful assistant. 请用山西话表达。<|endofprompt|>", + "You are a helpful assistant. 请用陕西话表达。<|endofprompt|>", + "You are a helpful assistant. 请用山东话表达。<|endofprompt|>", + "You are a helpful assistant. 请用上海话表达。<|endofprompt|>", + "You are a helpful assistant. 请用四川话表达。<|endofprompt|>", + "You are a helpful assistant. 请用天津话表达。<|endofprompt|>", + "You are a helpful assistant. 请用云南话表达。<|endofprompt|>", + "You are a helpful assistant. Please say a sentence as loudly as possible.<|endofprompt|>", + "You are a helpful assistant. Please say a sentence in a very soft voice.<|endofprompt|>", + "You are a helpful assistant. 请用尽可能慢地语速说一句话。<|endofprompt|>", + "You are a helpful assistant. 请用尽可能快地语速说一句话。<|endofprompt|>", + "You are a helpful assistant. 请非常开心地说一句话。<|endofprompt|>", + "You are a helpful assistant. 请非常伤心地说一句话。<|endofprompt|>", + "You are a helpful assistant. 请非常生气地说一句话。<|endofprompt|>", + "You are a helpful assistant. 我想体验一下小猪佩奇风格，可以吗？<|endofprompt|>", + "You are a helpful assistant. 你可以尝试用机器人的方式解答吗？<|endofprompt|>"] def pad_list(xs: List[torch.Tensor], pad_value: int):