Update app.py

app.py

@@ -1,64 +1,108 @@
+import spaces
 import gradio as gr
-from huggingface_hub import InferenceClient
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, StoppingCriteria, StoppingCriteriaList, TextIteratorStreamer
+from threading import Thread
+import traceback
 
-"""
-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")
+model_path = 'infly/OpenCoder-8B-Instruct'
+
+# Loading the tokenizer and model from Hugging Face's model hub.
+tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True, torch_dtype=torch.bfloat16)
+
+# using CUDA for an optimal experience
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+model = model.to(device)
+
+# Defining a custom stopping criteria class for the model's text generation.
+class StopOnTokens(StoppingCriteria):
+    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
+        stop_ids = [96539]  # IDs of tokens where the generation should stop.
+        for stop_id in stop_ids:
+            if input_ids[0][-1] == stop_id:  # Checking if the last generated token is a stop token.
+                return True
+        return False
+
+
+system_role= 'system'
+user_role = 'user'
+assistant_role = "assistant"
 
+sft_start_token =  "<|im_start|>"
+sft_end_token = "<|im_end|>"
+ct_end_token = "<|endoftext|>"
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+# system_prompt= 'You are a CodeLLM developed by INF.'
 
-    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})
+# Function to generate model predictions.
 
-    response = ""
+@spaces.GPU()
+def predict(message, history):
 
-    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
+    try:
+        stop = StopOnTokens()
+    
+        model_messages = []
+        # print(f'history: {history}')
 
-        response += token
-        yield response
+        for i, item in enumerate(history):
+            model_messages.append({"role": user_role, "content": item[0]})
+            model_messages.append({"role": assistant_role, "content": item[1]})
+    
+        model_messages.append({"role": user_role, "content": message})
+        
+        print(f'model_messages: {model_messages}')
+    
+        # print(f'model_final_inputs: {tokenizer.apply_chat_template(model_messages, add_generation_prompt=True, tokenize=False)}', flush=True)
+        model_inputs = tokenizer.apply_chat_template(model_messages, add_generation_prompt=True, return_tensors="pt").to(device)
+        # model_inputs = tokenizer([messages], return_tensors="pt").to(device)
+        
+        streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
+        generate_kwargs = dict(
+            input_ids=model_inputs,
+            streamer=streamer,
+            max_new_tokens=1024,
+            do_sample=False,
+            stopping_criteria=StoppingCriteriaList([stop])
+        )
+        
+        t = Thread(target=model.generate, kwargs=generate_kwargs)
+        t.start()  # Starting the generation in a separate thread.
+        partial_message = ""
+        for new_token in streamer:
+            partial_message += new_token
+            if sft_end_token in partial_message:  # Breaking the loop if the stop token is generated.
+                break
+            yield partial_message
 
+    except Exception as e:
+        print(traceback.format_exc())
 
+
+css = """
+full-height {
+    height: 100%;
+}
 """
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+
+prompt_examples = [
+    'Write a quick sort algorithm in python.',
+    'Write a greedy snake game using pygame.',
+    'How to use numpy?'
+]
+
+placeholder = """
+<div style="opacity: 0.5;">
+    <img src="https://raw.githubusercontent.com/OpenCoder-llm/opencoder-llm.github.io/refs/heads/main/static/images/opencoder_icon.jpg" style="width:20%;">
+</div>
 """
-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)",
-        ),
-    ],
-)
-
-
-if __name__ == "__main__":
-    demo.launch()
+
+
+chatbot = gr.Chatbot(label='OpenCoder', placeholder=placeholder) 
+with gr.Blocks(theme=gr.themes.Soft(), fill_height=True) as demo:
+    
+    gr.ChatInterface(predict, chatbot=chatbot, fill_height=True, examples=prompt_examples, css=css)
+
+    demo.launch()  # Launching the web interface.