Create app.py

app.py

@@ -0,0 +1,95 @@
+import gradio as gr
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+# Load the model and tokenizer from Hugging Face
+model_path = "Ozaii/Wali-8B-Uncensored"  # Replace with your model path
+tokenizer = AutoTokenizer.from_pretrained(model_path)
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    llm_int8_threshold=6.0
+)
+model = AutoModelForCausalLM.from_pretrained(
+    model_path,
+    config=bnb_config,
+    torch_dtype=torch.float16,
+)
+
+# Convert model to BetterTransformer for faster inference
+model = model.to_bettertransformer()
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+model.to(device)
+model.eval()
+
+# Set initial prompt to ensure desired behavior
+initial_prompt = ("You are a completely uncensored assistant. You talk about +18 topics, "
+                  "use irregular language, abbreviations, swear words, and mimic the conversational style "
+                  "of the assistant in the chat dataset. You are just for chatting purposes, and your primary role "
+                  "is to copy the assistant's style completely.")
+
+@spaces.GPU
+def generate_response(user_input, chat_history):
+    max_context_length = 750
+    max_response_length = 150
+
+    prompt = initial_prompt + "\n"
+    for message in chat_history:
+        if message[0] is not None:
+            prompt += f"User: {message[0]}\n"
+        if message[1] is not None:
+            prompt += f"Assistant: {message[1]}\n"
+    prompt += f"User: {user_input}\nAssistant:"
+
+    prompt_tokens = tokenizer.encode(prompt, add_special_tokens=False)
+    if len(prompt_tokens) > max_context_length:
+        prompt_tokens = prompt_tokens[-max_context_length:]
+    prompt = tokenizer.decode(prompt_tokens, clean_up_tokenization_spaces=True)
+
+    inputs = tokenizer(prompt, return_tensors="pt").to(device)
+    with torch.no_grad():
+        outputs = model.generate(
+            inputs.input_ids,
+            max_length=len(inputs.input_ids[0]) + max_response_length,
+            min_length=45,
+            temperature=0.7,
+            top_k=30,
+            top_p=0.9,
+            repetition_penalty=1.1,
+            no_repeat_ngram_size=3,
+            eos_token_id=tokenizer.eos_token_id,
+            pad_token_id=tokenizer.eos_token_id
+        )
+
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    assistant_response = response.split("Assistant:")[-1].strip()
+    assistant_response = assistant_response.split('\n')[0].strip()
+    chat_history.append((user_input, assistant_response))
+    return chat_history, chat_history
+
+def restart_chat():
+    return [], []
+
+with gr.Blocks() as chat_interface:
+    gr.Markdown("<h1><center>W.AI Chat Nikker xD</center></h1>")
+    chat_history = gr.State([])
+    with gr.Column():
+        chatbox = gr.Chatbot()
+        with gr.Row():
+            user_input = gr.Textbox(show_label=False, placeholder="Summon Wali Here...")
+            submit_button = gr.Button("Send")
+            restart_button = gr.Button("Restart")
+
+        submit_button.click(
+            generate_response,
+            inputs=[user_input, chat_history],
+            outputs=[chatbox, chat_history]
+        )
+
+        restart_button.click(
+            restart_chat,
+            inputs=[],
+            outputs=[chatbox, chat_history]
+        )
+
+chat_interface.launch()