thomas-yanxin commited on
Commit
dc18ab0
1 Parent(s): ed22028
Files changed (2) hide show
  1. app.py +0 -1
  2. chatllm.py +1 -1
app.py CHANGED
@@ -26,7 +26,6 @@ embedding_model_dict = {
26
 
27
  llm_model_dict = {
28
  "ChatGLM-6B-int4": "THUDM/chatglm-6b-int4",
29
- "ChatGLM-6B-int8": "THUDM/chatglm-6b-int8",
30
  "ChatGLM-6b-int4-qe": "THUDM/chatglm-6b-int4-qe",
31
  "Minimax": "Minimax"
32
  }
 
26
 
27
  llm_model_dict = {
28
  "ChatGLM-6B-int4": "THUDM/chatglm-6b-int4",
 
29
  "ChatGLM-6b-int4-qe": "THUDM/chatglm-6b-int4-qe",
30
  "Minimax": "Minimax"
31
  }
chatllm.py CHANGED
@@ -9,7 +9,7 @@ from transformers import AutoModel, AutoTokenizer
9
 
10
  os.environ["TOKENIZERS_PARALLELISM"] = "false"
11
 
12
- DEVICE = "cpu"
13
  DEVICE_ID = "0"
14
  CUDA_DEVICE = f"{DEVICE}:{DEVICE_ID}" if DEVICE_ID else DEVICE
15
 
 
9
 
10
  os.environ["TOKENIZERS_PARALLELISM"] = "false"
11
 
12
+ DEVICE = "cuda"
13
  DEVICE_ID = "0"
14
  CUDA_DEVICE = f"{DEVICE}:{DEVICE_ID}" if DEVICE_ID else DEVICE
15