oleksandrfluxon commited on
Commit
73697bb
·
1 Parent(s): 351fc66

use static path oleksandrfluxon/mpt-7b-chat-4bit in custom handler

Browse files
Files changed (1) hide show
  1. pipeline.py +6 -1
pipeline.py CHANGED
@@ -3,11 +3,13 @@ import transformers
3
  from typing import Dict, List, Any
4
 
5
  class PreTrainedPipeline():
6
- def __init__(self, path="oleksandrfluxon/mpt-7b-chat-4bit"):
 
7
  print("===> path", path)
8
  config = transformers.AutoConfig.from_pretrained(path, trust_remote_code=True)
9
  config.max_seq_len = 4096 # (input + output) tokens can now be up to 4096
10
 
 
11
  model = transformers.AutoModelForCausalLM.from_pretrained(
12
  path,
13
  config=config,
@@ -15,10 +17,12 @@ class PreTrainedPipeline():
15
  trust_remote_code=True,
16
  load_in_4bit=True, # Load model in the lowest 4-bit precision quantization
17
  )
 
18
 
19
  tokenizer = transformers.AutoTokenizer.from_pretrained('EleutherAI/gpt-neox-20b', padding_side="left", device_map="auto")
20
 
21
  self.pipeline = transformers.pipeline('text-generation', model=model, tokenizer=tokenizer)
 
22
 
23
  def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
24
  """
@@ -36,5 +40,6 @@ class PreTrainedPipeline():
36
  print("===> inputs", parameters)
37
 
38
  result = self.pipeline(inputs, **parameters)
 
39
 
40
  return result
 
3
  from typing import Dict, List, Any
4
 
5
  class PreTrainedPipeline():
6
+ def __init__(self, path=""):
7
+ path = "oleksandrfluxon/mpt-7b-chat-4bit"
8
  print("===> path", path)
9
  config = transformers.AutoConfig.from_pretrained(path, trust_remote_code=True)
10
  config.max_seq_len = 4096 # (input + output) tokens can now be up to 4096
11
 
12
+ print("===> loading model")
13
  model = transformers.AutoModelForCausalLM.from_pretrained(
14
  path,
15
  config=config,
 
17
  trust_remote_code=True,
18
  load_in_4bit=True, # Load model in the lowest 4-bit precision quantization
19
  )
20
+ print("===> model loaded")
21
 
22
  tokenizer = transformers.AutoTokenizer.from_pretrained('EleutherAI/gpt-neox-20b', padding_side="left", device_map="auto")
23
 
24
  self.pipeline = transformers.pipeline('text-generation', model=model, tokenizer=tokenizer)
25
+ print("===> init finished")
26
 
27
  def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
28
  """
 
40
  print("===> inputs", parameters)
41
 
42
  result = self.pipeline(inputs, **parameters)
43
+ print("===> result", result)
44
 
45
  return result