StudioIlios commited on
Commit
df40ff0
·
verified ·
1 Parent(s): c89f8cf

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +17 -6
app.py CHANGED
@@ -1,29 +1,40 @@
 
1
  import gradio as gr
2
  import torch
3
  from transformers import AutoTokenizer, AutoModelForCausalLM
4
  from peft import PeftModel
5
- import os
6
-
7
 
8
  hf_token = os.environ.get("HF_TOKEN")
9
 
 
10
  tokenizer = AutoTokenizer.from_pretrained("StudioIlios/icd10-model", token=hf_token)
 
 
11
  base_model = AutoModelForCausalLM.from_pretrained(
12
  "meta-llama/Llama-3.2-3B",
13
  token=hf_token,
14
- device_map="cpu",
15
- torch_dtype=torch.float32
 
16
  )
17
 
 
18
  model = PeftModel.from_pretrained(base_model, "StudioIlios/icd10-model", token=hf_token)
 
19
 
20
  def predict(clinical_text):
21
  inputs = tokenizer(clinical_text, return_tensors="pt")
22
- outputs = model.generate(**inputs, max_new_tokens=20, temperature=0.1)
 
23
  response = tokenizer.decode(outputs[0], skip_special_tokens=True)
24
  return response.replace(clinical_text, "").strip()
25
 
26
- demo = gr.Interface(fn=predict, inputs="text", outputs="text")
 
 
 
 
 
27
 
28
  # REST API at /api/predict
29
  demo.launch()
 
1
+ import os
2
  import gradio as gr
3
  import torch
4
  from transformers import AutoTokenizer, AutoModelForCausalLM
5
  from peft import PeftModel
 
 
6
 
7
  hf_token = os.environ.get("HF_TOKEN")
8
 
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",
15
  token=hf_token,
16
+ low_cpu_mem_usage=True,
17
+ torch_dtype=torch.float16,
18
+ device_map="cpu"
19
  )
20
 
21
+ print("Attaching LoRA adapter...")
22
  model = PeftModel.from_pretrained(base_model, "StudioIlios/icd10-model", token=hf_token)
23
+ print("Model loaded successfully!")
24
 
25
  def predict(clinical_text):
26
  inputs = tokenizer(clinical_text, return_tensors="pt")
27
+ with torch.no_grad():
28
+ outputs = model.generate(**inputs, max_new_tokens=30, temperature=0.1)
29
  response = tokenizer.decode(outputs[0], skip_special_tokens=True)
30
  return response.replace(clinical_text, "").strip()
31
 
32
+ demo = gr.Interface(
33
+ fn=predict,
34
+ inputs=gr.Textbox(placeholder="Enter symptom description here..."),
35
+ outputs="text",
36
+ title="ICD-10 Predictor API"
37
+ )
38
 
39
  # REST API at /api/predict
40
  demo.launch()