Deploy latest SLM customer-support chatbot

app.py

@@ -1,64 +1,10 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
-
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
-
-
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
-
-        response += token
-        yield response
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
-
+from module import chat_with_memory
 
+with gr.Blocks() as demo:
+    chatbot = gr.Chatbot()
+    txt     = gr.Textbox(placeholder="Type your message and press ⏎")
+    txt.submit(lambda msg, hist: (None, hist + [[msg, chat_with_memory(msg)]]),
+               [txt, chatbot], [txt, chatbot])
 if __name__ == "__main__":
     demo.launch()

module.py

@@ -0,0 +1,108 @@
+import re
+from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+from langchain.memory import ConversationBufferMemory
+import torch, unsloth, triton
+
+# ─── LOAD MODEL & TOKENIZER ─────────────────────────────────────────────
+# Adjust paths or HF repo as needed
+FINETUNED_DIR = "/content/drive/MyDrive/bitext-qlora-tinyllama"
+bnb_cfg = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_compute_dtype=torch.bfloat16
+)
+tokenizer = AutoTokenizer.from_pretrained(FINETUNED_DIR, use_fast=False)
+tokenizer.pad_token_id    = tokenizer.eos_token_id
+tokenizer.padding_side    = "left"
+tokenizer.truncation_side = "right"
+model = AutoModelForCausalLM.from_pretrained(
+    FINETUNED_DIR,
+    quantization_config=bnb_cfg,
+    device_map="auto",
+    trust_remote_code=True
+)
+
+# ─── MEMORY & PIPELINE ─────────────────────────────────────────────────
+memory         = ConversationBufferMemory(memory_key="user_lines",
+                                          human_prefix="User",
+                                          ai_prefix="Assistant",
+                                          return_messages=False)
+stored_order   = None
+pending_intent = None
+
+chat_pipe = pipeline(
+    "text-generation",
+    model=model,
+    tokenizer=tokenizer,
+    trust_remote_code=True,
+    return_full_text=False
+)
+
+# ─── HELPERS & HANDLERS ────────────────────────────────────────────────
+order_re = re.compile(r"#(\\d{1,10})")
+def extract_order(text):
+    m = order_re.search(text)
+    return m.group(1) if m else None
+
+def handle_status(o):
+    return f"Order #{o} is in transit and should arrive in 3–5 business days."
+def handle_eta(o):
+    return (f"Delivery for order #{o} typically takes 3–5 days; "
+            f"you can track it at https://track.example.com/{o}")
+def handle_track(o):
+    return f"Track order #{o} here: https://track.example.com/{o}"
+def handle_link(o):
+    return f"Here’s the latest tracking link for order #{o}: https://track.example.com/{o}"
+def handle_return_policy(_=None):
+    return ("Our return policy allows returns of unused items in their original packaging "
+            "within 30 days of receipt. Would you like me to connect you with a human agent?")
+def handle_gratitude(_=None):
+    return "You’re welcome! Is there anything else I can help with?"
+def handle_escalation(_=None):
+    return "I’m sorry, I don’t have that information. Would you like me to connect you with a human agent?"
+
+# ─── MAIN CHAT FUNCTION ────────────────────────────────────────────────
+def chat_with_memory(user_input: str) -> str:
+    global stored_order, pending_intent
+
+    memory.save_context({"input": user_input}, {"output": ""})
+    new_o = extract_order(user_input)
+    if new_o:
+        stored_order = new_o
+        if pending_intent in ("status","eta","track","link"):
+            fn = {"status":handle_status,"eta":handle_eta,
+                  "track":handle_track,"link":handle_link}[pending_intent]
+            reply = fn(stored_order)
+            pending_intent = None
+            memory.save_context({"input": user_input}, {"output": reply})
+            return reply
+
+    ui = user_input.lower().strip()
+    if any(tok in ui for tok in ["thank you","thanks","thx"]):
+        reply = handle_gratitude()
+    elif "return" in ui:
+        reply = handle_return_policy()
+    elif any(k in ui for k in ["status","where is my order","check status"]):
+        intent = "status"
+    elif any(k in ui for k in ["how long","eta","delivery time"]):
+        intent = "eta"
+    elif any(k in ui for k in ["how can i track","track my order","where is my package"]):
+        intent = "track"
+    elif "tracking link" in ui or "resend" in ui:
+        intent = "link"
+    else:
+        intent = "fallback"
+
+    if intent in ("status","eta","track","link"):
+        if not stored_order:
+            pending_intent = intent
+            reply = "Sure—what’s your order number (e.g., #12345)?"
+        else:
+            reply = {"status":handle_status,"eta":handle_eta,
+                     "track":handle_track,"link":handle_link}[intent](stored_order)
+    else:
+        reply = handle_escalation()
+
+    memory.save_context({"input": user_input}, {"output": reply})
+    return reply

requirements.txt

@@ -1 +1,10 @@
-huggingface_hub==0.25.2
+gradio
+transformers
+bitsandbytes
+accelerate
+xformers
+sentencepiece
+langchain
+unsloth
+unsloth_zoo
+huggingface_hub