Petro commited on
Commit
b0a8dfd
·
1 Parent(s): 48afe02

First model version

Browse files
Files changed (2) hide show
  1. main.py +12 -3
  2. requirements.txt +3 -1
main.py CHANGED
@@ -1,17 +1,26 @@
 
 
1
  from langchain.llms import CTransformers
2
  from fastapi import FastAPI
3
  from pydantic import BaseModel
4
 
5
  file_name = "zephyr-7b-beta.Q4_K_S.gguf"
6
  config = {
7
- "max_new_tokens": 1024,
8
- "model_type": "mistral",
9
- # "stream": True,
 
 
 
 
10
  }
11
  llm = CTransformers(
12
  model=file_name,
 
 
13
  **config
14
  )
 
15
 
16
 
17
  class validation(BaseModel):
 
1
+ import os
2
+
3
  from langchain.llms import CTransformers
4
  from fastapi import FastAPI
5
  from pydantic import BaseModel
6
 
7
  file_name = "zephyr-7b-beta.Q4_K_S.gguf"
8
  config = {
9
+ "max_new_token": 1024,
10
+ "repetition_penalty": 1.1,
11
+ "temperature": 0.5,
12
+ "top_k": 50,
13
+ "top_p": 0.9,
14
+ "stream": True,
15
+ "threads": int(os.cpu_count() / 2),
16
  }
17
  llm = CTransformers(
18
  model=file_name,
19
+ model_type="mistral",
20
+ lib="avx2",
21
  **config
22
  )
23
+ print(llm)
24
 
25
 
26
  class validation(BaseModel):
requirements.txt CHANGED
@@ -5,4 +5,6 @@ uvicorn
5
  requests
6
  python-dotenv
7
  ctransformers
8
- langchain
 
 
 
5
  requests
6
  python-dotenv
7
  ctransformers
8
+ torch
9
+ langchain
10
+ ctransformers