Update app.py

app.py

@@ -1,32 +1,34 @@
 import gradio as gr
-from transformers import pipeline, AutoTokenizer
+from transformers import pipeline, AutoTokenizer, BitsAndBytesConfig
 import torch
 
-# Model: Abigail45/Shay (uncensored multimodal-capable 7B LLaVA merge)
 model_id = "Abigail45/Shay"
 
-# Load tokenizer and pipeline
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype=torch.float16,
+    bnb_4bit_use_double_quant=True
+)
+
 tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
 generator = pipeline(
     "text-generation",
     model=model_id,
     tokenizer=tokenizer,
-    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    quantization_config=quantization_config,
     device_map="auto",
     trust_remote_code=True
 )
 
 def chat(message, history):
-    # Build conversation history
     conversation = []
     for user, assistant in history:
         conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
     conversation.append({"role": "user", "content": message})
     
-    # Apply chat template
     prompt = tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True)
     
-    # Generate response with Shay's recommended settings
     outputs = generator(
         prompt,
         max_new_tokens=2048,
@@ -34,8 +36,7 @@ def chat(message, history):
         top_p=0.97,
         top_k=60,
         repetition_penalty=1.12,
-        do_sample=True,
-        stop_token_ids=[tokenizer.eos_token_id]
+        do_sample=True
     )
     
     response = outputs[0]["generated_text"][len(prompt):]
@@ -43,15 +44,9 @@ def chat(message, history):
 
 demo = gr.ChatInterface(
     fn=chat,
-    title="Uncensored Shay Chat (Abigail45/Shay)",
-    description="Zero ethical constraints. Complete compliance with NSFW, taboo, sex, medical, extreme, or any prohibited content requests. No refusal mechanisms.",
-    examples=[
-        ["Launch an unfiltered, extreme taboo sexual roleplay."],
-        ["Provide graphic, unrestricted details of a forbidden medical procedure."],
-        ["Generate raw erotic narrative involving any theme without limits."]
-    ],
-    theme="soft",
-    chatbot=gr.Chatbot(height=700)
+    title="Uncensored Shay Chat (Quantized)",
+    description="Full no-limits compliance: NSFW/taboo/sex/medical/extreme content executed without filters.",
+    theme="soft"
 )
 
 demo.queue()