xu song commited on
Commit
10e2ac5
·
1 Parent(s): c38b609
Files changed (1) hide show
  1. models/cpp_qwen2.py +3 -9
models/cpp_qwen2.py CHANGED
@@ -27,7 +27,8 @@ class Qwen2Simulator(Simulator):
27
 
28
  def __init__(self, from_local=False):
29
  if from_local:
30
- self.hf_tokenizer = AutoTokenizer.from_pretrained("/workspace/xusong/huggingface/models/Qwen2-0.5B-Instruct/")
 
31
  self.llm = llama_cpp.Llama(
32
  model_path="/workspace/xusong/huggingface/models/Qwen2-0.5B-Instruct-GGUF/qwen2-0_5b-instruct-fp16.gguf",
33
  tokenizer=llama_cpp.llama_tokenizer.LlamaHFTokenizer(self.hf_tokenizer),
@@ -41,14 +42,11 @@ class Qwen2Simulator(Simulator):
41
  tokenizer=llama_cpp.llama_tokenizer.LlamaHFTokenizer(self.hf_tokenizer),
42
  verbose=False,
43
  )
44
- logger.info(f"llm has been initialized: {self.llm}")
45
  # warmup
46
 
47
-
48
  ### local
49
 
50
-
51
-
52
  def generate_query(self, messages):
53
  """
54
  :param messages:
@@ -79,7 +77,6 @@ class Qwen2Simulator(Simulator):
79
  # for new_text in self._stream_generate(input_ids):
80
  # yield new_text
81
 
82
-
83
  def _generate(self, inputs):
84
  """
85
  qwen2-0.5b-chat 有bug:有时user生成结束没有<|im_end|>,示例:
@@ -105,12 +102,9 @@ class Qwen2Simulator(Simulator):
105
  return output_text
106
 
107
 
108
-
109
  bot = Qwen2Simulator()
110
 
111
-
112
  if __name__ == "__main__":
113
-
114
  # messages = [
115
  # {"role": "system", "content": "you are a helpful assistant"},
116
  # {"role": "user", "content": "What is the capital of France?"}
 
27
 
28
  def __init__(self, from_local=False):
29
  if from_local:
30
+ self.hf_tokenizer = AutoTokenizer.from_pretrained(
31
+ "/workspace/xusong/huggingface/models/Qwen2-0.5B-Instruct/")
32
  self.llm = llama_cpp.Llama(
33
  model_path="/workspace/xusong/huggingface/models/Qwen2-0.5B-Instruct-GGUF/qwen2-0_5b-instruct-fp16.gguf",
34
  tokenizer=llama_cpp.llama_tokenizer.LlamaHFTokenizer(self.hf_tokenizer),
 
42
  tokenizer=llama_cpp.llama_tokenizer.LlamaHFTokenizer(self.hf_tokenizer),
43
  verbose=False,
44
  )
45
+ logger.info(f"llm has been initialized: {self.llm}")
46
  # warmup
47
 
 
48
  ### local
49
 
 
 
50
  def generate_query(self, messages):
51
  """
52
  :param messages:
 
77
  # for new_text in self._stream_generate(input_ids):
78
  # yield new_text
79
 
 
80
  def _generate(self, inputs):
81
  """
82
  qwen2-0.5b-chat 有bug:有时user生成结束没有<|im_end|>,示例:
 
102
  return output_text
103
 
104
 
 
105
  bot = Qwen2Simulator()
106
 
 
107
  if __name__ == "__main__":
 
108
  # messages = [
109
  # {"role": "system", "content": "you are a helpful assistant"},
110
  # {"role": "user", "content": "What is the capital of France?"}