JimmyTsang commited on
Commit
56467fc
·
verified ·
1 Parent(s): e35a637

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +11 -25
app.py CHANGED
@@ -8,33 +8,21 @@ from huggingface_hub import snapshot_download
8
  # Ensure your Hugging Face token is set
9
  TOKEN = os.getenv('HF_TOKEN')
10
 
11
- # Define paths and download the model files
12
- mistral_models_path = Path.home().joinpath('mistral_models', 'Meta-Llama-3.1-8B-Instruct')
13
- mistral_models_path.mkdir(parents=True, exist_ok=True)
14
 
15
- snapshot_download(repo_id="meta-llama/Meta-Llama-3.1-8B-Instruct",
16
- allow_patterns=["*.json", "*.safetensors", "*.model"],
17
- local_dir=mistral_models_path,
18
- token=TOKEN)
19
-
20
- # Load Meta-Llama-3.1-8B-Instruct model and tokenizer
21
- model_path = str(Path.home().joinpath('mistral_models', 'Meta-Llama-3.1-8B-Instruct'))
22
- llama_model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype=torch.bfloat16, device_map="auto")
23
- llama_tokenizer = AutoTokenizer.from_pretrained(model_path)
24
 
25
  # Define the respond function to generate test cases for a code solution
26
- def respond(code_problem, code_solution, max_tokens, temperature, top_p):
27
  prompt = f"Generate test cases to test the following code solution:\n\nProblem:\n{code_problem}\n\nSolution:\n{code_solution}\n\nTest Cases:"
28
- inputs = llama_tokenizer(prompt, return_tensors="pt").to("cuda")
29
- generated_ids = llama_model.generate(
30
- inputs["input_ids"],
31
- max_new_tokens=max_tokens,
32
- do_sample=True,
33
- temperature=temperature,
34
- top_p=top_p
35
- )
36
- result = llama_tokenizer.decode(generated_ids[0], skip_special_tokens=True)
37
- return result
38
 
39
  # Define Gradio interface
40
  interface = gr.Interface(
@@ -43,8 +31,6 @@ interface = gr.Interface(
43
  gr.Textbox(lines=10, placeholder="Describe the code problem here...", label="Code Problem"),
44
  gr.Textbox(lines=10, placeholder="Provide the code solution here...", label="Code Solution"),
45
  gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
46
- gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
47
- gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
48
  ],
49
  outputs=gr.Textbox(lines=20, label="Generated Test Cases"),
50
  title="Code Test Case Generator",
 
8
  # Ensure your Hugging Face token is set
9
  TOKEN = os.getenv('HF_TOKEN')
10
 
11
+ model_id = "meta-llama/Meta-Llama-3.1-8B-Instruct"
 
 
12
 
13
+ pipeline = transformers.pipeline(
14
+ "text-generation",
15
+ model=model_id,
16
+ model_kwargs={"torch_dtype": torch.bfloat16},
17
+ device_map="auto",
18
+ token=TOKEN
19
+ )
 
 
20
 
21
  # Define the respond function to generate test cases for a code solution
22
+ def respond(code_problem, code_solution, max_tokens):
23
  prompt = f"Generate test cases to test the following code solution:\n\nProblem:\n{code_problem}\n\nSolution:\n{code_solution}\n\nTest Cases:"
24
+ outputs = pipeline(prompt, max_new_tokens=max)
25
+ return outputs[0]["generated_text"][-1]
 
 
 
 
 
 
 
 
26
 
27
  # Define Gradio interface
28
  interface = gr.Interface(
 
31
  gr.Textbox(lines=10, placeholder="Describe the code problem here...", label="Code Problem"),
32
  gr.Textbox(lines=10, placeholder="Provide the code solution here...", label="Code Solution"),
33
  gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
 
 
34
  ],
35
  outputs=gr.Textbox(lines=20, label="Generated Test Cases"),
36
  title="Code Test Case Generator",