|
import gradio as gr |
|
from huggingfacehub import InferenceClient, HfApi |
|
import os |
|
import requests |
|
import pandas as pd |
|
import json |
|
|
|
|
|
hftoken = os.getenv("H") |
|
|
|
if not hftoken: |
|
raise ValueError("H νκ²½ λ³μκ° μ€μ λμ§ μμμ΅λλ€.") |
|
|
|
|
|
api = HfApi(token=hftoken) |
|
|
|
try: |
|
client = InferenceClient("meta-llama/Meta-Llama-3-70B-Instruct", token="H") |
|
except Exception as e: |
|
print(f"rror initializing InferenceClient: {e}") |
|
|
|
|
|
|
|
|
|
currentdir = os.path.dirname(os.path.abspath(file)) |
|
csvpath = os.path.join(currentdir, 'prompts.csv') |
|
datapath = os.path.join(currentdir, 'newdataset.parquet') |
|
|
|
|
|
promptsdf = pd.readcsv(csvpath) |
|
datadf = pd.readparquet(datapath) |
|
|
|
def getprompt(act): |
|
matchingprompt = promptsdf[promptsdf['act'] == act]['prompt'].values |
|
return matchingprompt[0] if len(matchingprompt) 0 else None |
|
|
|
def respond( |
|
message, |
|
history: list[tuple[str, str]], |
|
systemmessage, |
|
maxtokens, |
|
temperature, |
|
topp, |
|
): |
|
|
|
prompt = getprompt(message) |
|
if prompt: |
|
response = prompt |
|
else: |
|
systemprefix = """ |
|
λΉμ μ μ±λ΄μ
λλ€. λͺ¨λ μ§λ¬Έμ λν΄ μΉμ νκ³ μ νν λ΅λ³μ μ 곡νμΈμ. |
|
μ§λ¬Έμ λν λ΅λ³μ μ°Ύμ μ μλ κ²½μ°, μ μ ν λμμ μ κ³΅ν΄ μ£ΌμΈμ. |
|
""" |
|
|
|
fullprompt = f"{systemprefix} {systemmessage}\n\n" |
|
|
|
for user, assistant in history: |
|
fullprompt += f"Human: {user}\nAI: {assistant}\n" |
|
|
|
fullprompt += f"Human: {message}\nAI:" |
|
APIL = "https://api-inference.huggingface.co/models/meta-llama/Meta-Llama-3-70B-Instruct" |
|
headers = {"Authorization": f"Bearer {hftoken}"} |
|
|
|
def query(payload): |
|
response = requests.post(APIL, headers=headers, json=payload) |
|
return response.text |
|
|
|
try: |
|
payload = { |
|
"inputs": fullprompt, |
|
"parameters": { |
|
"maxnewtokens": maxtokens, |
|
"temperature": temperature, |
|
"topp": topp, |
|
"returnfulltext": False |
|
}, |
|
} |
|
rawresponse = query(payload) |
|
print("aw API response:", rawresponse) |
|
|
|
try: |
|
output = json.loads(rawresponse) |
|
if isinstance(output, list) and len(output) 0 and "generatedtext" in output[0]: |
|
response = output[0]["generatedtext"] |
|
else: |
|
response = f"μμμΉ λͺ»ν μλ΅ νμμ
λλ€: {output}" |
|
except json.JSecoderror: |
|
response = f"JS λμ½λ© μ€λ₯. μμ μλ΅: {rawresponse}" |
|
|
|
except Exception as e: |
|
print(f"rror during API request: {e}") |
|
response = f"μ£μ‘ν©λλ€. μλ΅ μμ± μ€ μ€λ₯κ° λ°μνμ΅λλ€: {str(e)}" |
|
|
|
yield response |
|
|
|
demo = gr.ChatInterface( |
|
respond, |
|
title="My Chatbot", |
|
description= "his is my chatbot!", |
|
additionalinputs=[ |
|
gr.extbox(value=""" |
|
λΉμ μ μ±λ΄μ
λλ€. λͺ¨λ μ§λ¬Έμ λν΄ μΉμ νκ³ μ νν λ΅λ³μ μ 곡νμΈμ. |
|
μ§λ¬Έμ λν λ΅λ³μ μ°Ύμ μ μλ κ²½μ°, μ μ ν λμμ μ κ³΅ν΄ μ£ΌμΈμ. |
|
""", label="μμ€ν
ν둬ννΈ"), |
|
gr.Slider(minimum=1, maximum=4000, value=2000, step=1, label="Max new tokens"), |
|
gr.Slider(minimum=0.1, maximum=4.0, value=1.0, step=0.1, label="temperature"), |
|
gr.Slider( |
|
minimum=0.1, |
|
maximum=1.0, |
|
value=0.95, |
|
step=0.05, |
|
label="top-p (nucleus sampling)", |
|
), |
|
], |
|
examples=[ |
|
["νκΈλ‘ λ΅λ³ν κ²"], |
|
["κ³μ μ΄μ΄μ μμ±νλΌ"], |
|
], |
|
cacheexamples=alse, |
|
) |
|
|
|
if name == "main": |
|
demo.launch() |