Merge pull request #455 from QwenLM/logicwong-patch-5

Update README.md, update batch infer

README.md

@@ -376,15 +376,16 @@ import torch
 from tokenization_qwen import QWenTokenizer
 from modeling_qwen import QWenLMHeadModel
 from transformers import GenerationConfig
-from qwen_generation_utils import make_context
+from qwen_generation_utils import make_context, decode_tokens, get_stop_words_ids
 
 
 tokenizer = QWenTokenizer.from_pretrained('./', pad_token='<|extra_0|>', eos_token='<|endoftext|>', padding_side='left')
 model = QWenLMHeadModel.from_pretrained('./', device_map="auto").eval()
 model.generation_config = GenerationConfig.from_pretrained('./')
+stop_words_ids = get_stop_words_ids(model.generation_config.chat_format, tokenizer)
 
 all_raw_text = ["我想听你说爱我。", "今天我想吃点啥，甜甜的，推荐下", "我马上迟到了，怎么做才能不迟到"]
-batch_question = []
+batch_raw_text = []
 for q in all_raw_text:
     raw_text, _ = make_context(
         tokenizer,
@@ -393,17 +394,29 @@ for q in all_raw_text:
         max_window_size=model.generation_config.max_window_size,
         chat_format=model.generation_config.chat_format,
     )
-    batch_question.append(raw_text)
+    batch_raw_text.append(raw_text)
 
-batch_input_ids = tokenizer(batch_question, padding='longest')
-print(batch_input_ids)
-
-batch_input_ids1 = torch.LongTensor(batch_input_ids['input_ids']).to(model.device)
+batch_input_ids = tokenizer(batch_raw_text, padding='longest')
+batch_input_ids = torch.LongTensor(batch_input_ids['input_ids']).to(model.device)
 batch_out_ids = model.generate(
-    input_ids=batch_input_ids1
-    ,return_dict_in_generate=False
+    batch_input_ids,
+    stop_words_ids=stop_words_ids,
+    return_dict_in_generate=False,
+    generation_config=model.generation_config
 )
-batch_response = [tokenizer.decode(o, skip_special_tokens=True) for o in batch_out_ids]
+padding_lens = [batch_input_ids[i].eq(tokenizer.pad_token_id).sum().item() for i in range(batch_input_ids.size(0))]
+
+batch_response = [
+    decode_tokens(
+        batch_out_ids[i][padding_lens[i]:],
+        tokenizer,
+        raw_text_len=len(batch_raw_text[i]),
+        context_length=batch_input_ids[i].size(0),
+        chat_format="chatml",
+        verbose=False,
+        errors='replace'
+    ) for i in range(len(all_raw_text))
+]
 print(batch_response)
 
 response, _ = model.chat(tokenizer, "我想听你说爱我。", history=None)