StudioIlios commited on
Commit
4db90ae
·
verified ·
1 Parent(s): 597c7b0

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +10 -6
app.py CHANGED
@@ -9,6 +9,9 @@ hf_token = os.environ.get("HF_TOKEN")
9
  print("Loading tokenizer...")
10
  tokenizer = AutoTokenizer.from_pretrained("StudioIlios/icd10-model", token=hf_token)
11
 
 
 
 
12
  print("Loading base model in float16 precision...")
13
  base_model = AutoModelForCausalLM.from_pretrained(
14
  "meta-llama/Llama-3.2-3B",
@@ -23,15 +26,18 @@ model = PeftModel.from_pretrained(base_model, "StudioIlios/icd10-model", token=h
23
  print("Model loaded successfully!")
24
 
25
  def predict(clinical_text):
26
- inputs = tokenizer(clinical_text, return_tensors="pt")
 
 
27
  input_token_length = inputs["input_ids"].shape[1]
28
 
29
  with torch.no_grad():
30
  outputs = model.generate(
31
  **inputs,
32
- max_new_tokens=50,
33
- temperature=0.7,
34
- do_sample=True
 
35
  )
36
 
37
  generated_tokens = outputs[0][input_token_length:]
@@ -46,6 +52,4 @@ demo = gr.Interface(
46
  title="ICD-10 Predictor API"
47
  )
48
 
49
-
50
- # REST API at /api/predict
51
  demo.launch()
 
9
  print("Loading tokenizer...")
10
  tokenizer = AutoTokenizer.from_pretrained("StudioIlios/icd10-model", token=hf_token)
11
 
12
+ if tokenizer.pad_token_id is None:
13
+ tokenizer.pad_token_id = tokenizer.eos_token_id
14
+
15
  print("Loading base model in float16 precision...")
16
  base_model = AutoModelForCausalLM.from_pretrained(
17
  "meta-llama/Llama-3.2-3B",
 
26
  print("Model loaded successfully!")
27
 
28
  def predict(clinical_text):
29
+ prompt = f"{clinical_text}\n"
30
+ inputs = tokenizer(prompt, return_tensors="pt")
31
+
32
  input_token_length = inputs["input_ids"].shape[1]
33
 
34
  with torch.no_grad():
35
  outputs = model.generate(
36
  **inputs,
37
+ max_new_tokens=30,
38
+ do_sample=False,
39
+ repetition_penalty=1.15,
40
+ pad_token_id=tokenizer.eos_token_id
41
  )
42
 
43
  generated_tokens = outputs[0][input_token_length:]
 
52
  title="ICD-10 Predictor API"
53
  )
54
 
 
 
55
  demo.launch()