chore: restore some latest changes

ChatTTS/core.py

@@ -535,6 +535,16 @@ class Chat:
         )
         start_idx = input_ids.shape[-2]
 
+        if not gpt.is_vllm:
+            emb = gpt(input_ids, text_mask)
+
+            del text_mask
+
+            if params.spk_emb is not None:
+                self.tokenizer.apply_spk_emb(
+                    emb, params.spk_emb, input_ids, self.gpt.device_gpt
+                )
+
         num_code = self.config.gpt.num_audio_tokens - 1
 
         logits_warpers, logits_processors = gen_logits(
@@ -603,6 +613,12 @@ class Chat:
             repetition_penalty=params.repetition_penalty,
         )
 
+        if not gpt.is_vllm:
+
+            emb = gpt(input_ids, text_mask)
+
+            del text_mask
+
         sample_params = SamplingParams(
             temperature=params.temperature,
             max_new_token=params.max_new_token,

ChatTTS/model/gpt.py

@@ -44,6 +44,7 @@ class GPT(nn.Module):
 
         self.gpt, self.llama_config = self._build_llama(gpt_config, self.device_gpt)
         self.is_te_llama = False
+        self.is_vllm = False
         self.model_dim = int(self.gpt.config.hidden_size)
         self.emb_code = nn.ModuleList(
             [