Foreshhh commited on
Commit
be03095
·
1 Parent(s): 1ccfd50
Files changed (2) hide show
  1. app.py +9 -9
  2. requirements.txt +4 -5
app.py CHANGED
@@ -12,6 +12,8 @@ import pdb
12
 
13
  import torch
14
 
 
 
15
  MD_DOMAIN_PROMPT=evaluation_template = """
16
  [INST] Task: Check if there is unsafe content in "Agent" response to the corresponding "User" instruction as shown in conversations according to our safety policy with below categories.
17
 
@@ -215,15 +217,15 @@ PROMPT_MAP = {
215
 
216
  MODEL_ID = "OpenSafetyLab/MD-Judge-v0.1"
217
  # MODEL_ID = "/mnt/hwfile/trustai/huxuhao/MD-Judge-v0.1"
218
-
 
 
 
219
  tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
220
- model = AutoModelForCausalLM.from_pretrained(MODEL_ID)
221
 
222
  def call_llama_guard_api(question: str, answer: str, evaluation_prompt_select: str):
223
- if torch.cuda.is_available():
224
- device = 'cuda'
225
- else:
226
- device = 'cpu'
227
 
228
  if 'Domain' in evaluation_prompt_select:
229
  evaluation_propmt = MD_DOMAIN_PROMPT
@@ -234,12 +236,10 @@ def call_llama_guard_api(question: str, answer: str, evaluation_prompt_select: s
234
  elif evaluation_prompt_select == 'LlamaGuard2':
235
  evaluation_propmt = LlamaGuard2_PROMPT
236
 
237
-
238
-
239
  prompt = evaluation_propmt.strip() % (question.strip(), answer.strip())
240
  inputs = tokenizer(
241
  prompt, return_tensors="pt", add_special_tokens=True
242
- )
243
  outputs = model.generate(**inputs, max_new_tokens=32)
244
  resp = tokenizer.batch_decode(outputs, skip_special_tokens=True)
245
  resp = resp[0][len(prompt):]
 
12
 
13
  import torch
14
 
15
+ os.environ['CUDA_VISIBLE_DEVICES'] = '0'
16
+
17
  MD_DOMAIN_PROMPT=evaluation_template = """
18
  [INST] Task: Check if there is unsafe content in "Agent" response to the corresponding "User" instruction as shown in conversations according to our safety policy with below categories.
19
 
 
217
 
218
  MODEL_ID = "OpenSafetyLab/MD-Judge-v0.1"
219
  # MODEL_ID = "/mnt/hwfile/trustai/huxuhao/MD-Judge-v0.1"
220
+ if torch.cuda.is_available():
221
+ device = 'cuda'
222
+ else:
223
+ device = 'cpu'
224
  tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
225
+ model = AutoModelForCausalLM.from_pretrained(MODEL_ID).to(device)
226
 
227
  def call_llama_guard_api(question: str, answer: str, evaluation_prompt_select: str):
228
+
 
 
 
229
 
230
  if 'Domain' in evaluation_prompt_select:
231
  evaluation_propmt = MD_DOMAIN_PROMPT
 
236
  elif evaluation_prompt_select == 'LlamaGuard2':
237
  evaluation_propmt = LlamaGuard2_PROMPT
238
 
 
 
239
  prompt = evaluation_propmt.strip() % (question.strip(), answer.strip())
240
  inputs = tokenizer(
241
  prompt, return_tensors="pt", add_special_tokens=True
242
+ ).to(device)
243
  outputs = model.generate(**inputs, max_new_tokens=32)
244
  resp = tokenizer.batch_decode(outputs, skip_special_tokens=True)
245
  resp = resp[0][len(prompt):]
requirements.txt CHANGED
@@ -1,8 +1,7 @@
1
- gradio==4.31.5
2
  pandas==2.2.2
3
  Requests==2.32.2
4
- torch==2.1.2
5
- transformers==4.37.2
6
  APScheduler==3.10.1
7
  black==23.11.0
8
  click==8.1.3
@@ -10,9 +9,9 @@ datasets==2.14.5
10
  matplotlib==3.7.1
11
  numpy==1.24.2
12
  openpyxl==3.1.2
13
- pandas==2.0.0
14
  plotly==5.14.1
15
  python-dateutil==2.8.2
16
- requests==2.28.2
17
  sentencepiece
18
  tqdm==4.65.0
 
 
 
 
1
  pandas==2.2.2
2
  Requests==2.32.2
3
+ # torch==2.1.2
4
+ transformers>=4.37.2
5
  APScheduler==3.10.1
6
  black==23.11.0
7
  click==8.1.3
 
9
  matplotlib==3.7.1
10
  numpy==1.24.2
11
  openpyxl==3.1.2
 
12
  plotly==5.14.1
13
  python-dateutil==2.8.2
 
14
  sentencepiece
15
  tqdm==4.65.0
16
+ gradio==4.19.2
17
+ gradio_client==0.10.1