File size: 5,529 Bytes
0f5897f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 |
import os
import re
import logging
import textwrap
import autopep8
import gradio as gr
from huggingface_hub import hf_hub_download
from llama_cpp import Llama
import jwt
from typing import Generator
# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# JWT settings
JWT_SECRET = os.environ.get("JWT_SECRET")
JWT_ALGORITHM = "HS256"
# Model settings
MODEL_NAME = "leetmonkey_peft__q8_0.gguf"
REPO_ID = "sugiv/leetmonkey-peft-gguf"
# Generation parameters
generation_kwargs = {
"max_tokens": 2048,
"stop": ["```", "### Instruction:", "### Response:"],
"echo": False,
"temperature": 0.2,
"top_k": 50,
"top_p": 0.95,
"repeat_penalty": 1.1
}
def download_model(model_name: str) -> str:
logger.info(f"Downloading model: {model_name}")
model_path = hf_hub_download(
repo_id=REPO_ID,
filename=model_name,
cache_dir="./models",
force_download=True,
resume_download=True
)
logger.info(f"Model downloaded: {model_path}")
return model_path
# Download and load the 8-bit model at startup
model_path = download_model(MODEL_NAME)
llm = Llama(
model_path=model_path,
n_ctx=2048,
n_threads=4,
n_gpu_layers=-1, # Use all available GPU layers
verbose=False
)
logger.info("8-bit model loaded successfully")
def generate_solution(instruction: str) -> str:
system_prompt = "You are a Python coding assistant specialized in solving LeetCode problems. Provide only the complete implementation of the given function. Ensure proper indentation and formatting. Do not include any explanations or multiple solutions."
full_prompt = f"""### Instruction:
{system_prompt}
Implement the following function for the LeetCode problem:
{instruction}
### Response:
Here's the complete Python function implementation:
```python
"""
response = llm(full_prompt, **generation_kwargs)
return response["choices"][0]["text"]
def extract_and_format_code(text: str) -> str:
# Extract code between triple backticks
code_match = re.search(r'```python\s*(.*?)\s*```', text, re.DOTALL)
if code_match:
code = code_match.group(1)
else:
code = text
# Remove any text before the function definition
code = re.sub(r'^.*?(?=def\s+\w+\s*\()', '', code, flags=re.DOTALL)
# Dedent the code to remove any common leading whitespace
code = textwrap.dedent(code)
# Split the code into lines
lines = code.split('\n')
# Find the function definition line
func_def_index = next((i for i, line in enumerate(lines) if line.strip().startswith('def ')), 0)
# Ensure proper indentation
indented_lines = [lines[func_def_index]] # Keep the function definition as is
for line in lines[func_def_index + 1:]:
if line.strip(): # If the line is not empty
indented_lines.append(' ' + line) # Add 4 spaces of indentation
else:
indented_lines.append(line) # Keep empty lines as is
formatted_code = '\n'.join(indented_lines)
try:
return autopep8.fix_code(formatted_code)
except:
return formatted_code
def verify_token(token: str) -> bool:
try:
jwt.decode(token, JWT_SECRET, algorithms=[JWT_ALGORITHM])
return True
except jwt.PyJWTError:
return False
def generate_solution_api(instruction: str, token: str) -> str:
if not verify_token(token):
return "Invalid token. Please provide a valid JWT token."
logger.info("Generating solution")
generated_output = generate_solution(instruction)
formatted_code = extract_and_format_code(generated_output)
logger.info("Solution generated successfully")
return formatted_code
def stream_solution_api(instruction: str, token: str) -> Generator[str, None, None]:
if not verify_token(token):
yield "Invalid token. Please provide a valid JWT token."
return
logger.info("Streaming solution")
system_prompt = "You are a Python coding assistant specialized in solving LeetCode problems. Provide only the complete implementation of the given function. Ensure proper indentation and formatting. Do not include any explanations or multiple solutions."
full_prompt = f"""### Instruction:
{system_prompt}
Implement the following function for the LeetCode problem:
{instruction}
### Response:
Here's the complete Python function implementation:
```python
"""
generated_text = ""
for chunk in llm(full_prompt, stream=True, **generation_kwargs):
token = chunk["choices"]["text"]
generated_text += token
yield generated_text
formatted_code = extract_and_format_code(generated_text)
logger.info("Solution generated successfully")
yield formatted_code
# Gradio interface
def gradio_generate(instruction: str, token: str) -> str:
return generate_solution_api(instruction, token)
def gradio_stream(instruction: str, token: str) -> str:
return "".join(list(stream_solution_api(instruction, token)))
iface = gr.Interface(
fn=[gradio_generate, gradio_stream],
inputs=[
gr.Textbox(label="LeetCode Problem Instruction"),
gr.Textbox(label="JWT Token")
],
outputs=[
gr.Code(label="Generated Solution"),
gr.Code(label="Streamed Solution")
],
title="LeetCode Problem Solver",
description="Enter a LeetCode problem instruction and your JWT token to generate a solution."
)
if __name__ == "__main__":
iface.launch(share=True) |