main.py

from fastapi import FastAPI
from pydantic import BaseModel
from huggingface_hub import InferenceClient
import uvicorn

app = FastAPI()

# Initialize the InferenceClient with the specified model
client = InferenceClient("nvidia/Llama-3.1-Nemotron-70B-Instruct-HF")

# Define the structure of the request body
class CourseRequest(BaseModel):
    course_name: str
    history: list = []  # Keeping history optional
    temperature: float = 0.0
    max_new_tokens: int = 1048
    top_p: float = 0.15
    repetition_penalty: float = 1.0

# Format the prompt for the model
def format_prompt(course_name, history):
    prompt = "<s>"
    for user_prompt, bot_response in history:
        prompt += f"[INST] {user_prompt} [/INST] {bot_response} </s> "
    prompt += f"[INST] Generate a roadmap for the course: {course_name} [/INST]"
    return prompt

# Generate text using the specified parameters
def generate(course_request: CourseRequest):
    temperature = max(float(course_request.temperature), 1e-2)
    top_p = float(course_request.top_p)

    generate_kwargs = {
        'temperature': temperature,
        'max_new_tokens': course_request.max_new_tokens,
        'top_p': top_p,
        'repetition_penalty': course_request.repetition_penalty,
        'do_sample': True,
        'seed': 42,
    }

    formatted_prompt = format_prompt(course_request.course_name, course_request.history)
    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
    output = ""

    for response in stream:
        output += response.token.text
    return output

# Define the API endpoint for generating course roadmaps
@app.post("/generate-roadmap/")
async def generate_roadmap(course_request: CourseRequest):
    return {"roadmap": generate(course_request)}

# Run the application (uncomment the next two lines if running this as a standalone script)
# if __name__ == "__main__":
#     uvicorn.run(app, host="0.0.0.0", port=8000)