File size: 2,490 Bytes
e27c8af c90040d e27c8af c45dc39 c90040d c45dc39 c90040d c45dc39 e27c8af c90040d c45dc39 e27c8af c45dc39 c90040d c45dc39 c90040d e27c8af c45dc39 e27c8af c45dc39 e27c8af |
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 |
import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
from peft import PeftModel
# 1) Base model & tokenizer
BASE_MODEL = "facebook/blenderbot-400M-distill"
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
base_model = AutoModelForSeq2SeqLM.from_pretrained(BASE_MODEL)
# 2) Attach your LoRA adapter
ADAPTER_REPO = "abinashnp/bayedger-chatbot"
model = PeftModel.from_pretrained(base_model, ADAPTER_REPO)
# 3) Build the text2text pipeline (no explicit device arg)
chatbot = pipeline(
"text2text-generation",
model=model,
tokenizer=tokenizer,
# device_map="auto" # only if you use Accelerate; otherwise remove
)
# 4) System prompt (context) that always precedes user questions
SYSTEM_PROMPT = (
"You are BayEdger’s AI assistant. You only answer FAQs about BayEdger’s "
"services, pricing, and contact info. If you don’t know the answer, "
"you must say exactly:\n"
'"Sorry, I don’t have that info—please contact [email protected]."\n\n'
"Here is what you should know about BayEdger:\n"
"- AI‐powered websites and automation\n"
"- Chatbots, email agents, process automation, analytics, content gen\n"
"- Clear pricing tiers: Basic site ($400), Chatbot ($750+50/mo), Email ($1k+100/mo), etc.\n"
"- Starter/Growth/Premium bundles\n"
"- Contact: [email protected], +1‐234‐559‐87994, 13 Madison St, NY\n\n"
)
def respond(query):
# 5) Compose full prompt
prompt = (
SYSTEM_PROMPT
f"question: {query}\n"
"answer:"
)
# 6) Generate
out = chatbot(
prompt,
max_new_tokens=128,
do_sample=False,
num_beams=2,
early_stopping=True,
pad_token_id=tokenizer.eos_token_id
)[0]["generated_text"]
# 7) Strip off everything up through our "answer:" token
if "answer:" in out:
reply = out.split("answer:", 1)[1].strip()
else:
reply = out.strip()
# 8) Fallback: if the model didn’t produce anything substantial
if len(reply) < 15 or "don't know" in reply.lower() or "sorry" in reply.lower():
return "Sorry, I don’t have that info—please contact [email protected]."
return reply
# 9) Gradio UI
with gr.Blocks() as demo:
gr.Markdown("# 🤖 BayEdger FAQ Chatbot")
txt = gr.Textbox(placeholder="Ask me about BayEdger…", label="Your question")
out = gr.Textbox(label="Answer")
txt.submit(respond, txt, out)
demo.launch()
|