Update app.py

app.py

@@ -1,98 +1,104 @@
-import subprocess
-subprocess.run(
-    'pip install flash-attn --no-build-isolation',
-    env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"},
-    shell=True
-)
-
+import os
+import time
+import spaces
 import torch
-from PIL import Image
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
 import gradio as gr
-import spaces
-from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, StoppingCriteriaList, StoppingCriteria
-import os
 from threading import Thread
 
-
 HF_TOKEN = os.environ.get("HF_TOKEN", None)
-MODEL_LIST = "THUDM/LongWriter-glm4-9b"
-#MODELS = os.environ.get("MODELS")
-#MODEL_NAME = MODELS.split("/")[-1]
+MODEL = "LGAI-EXAONE/EXAONE-3.5-32B-Instruct"
 
-TITLE = "<h1><center>GLM SPACE</center></h1>"
+TITLE = "<h1><center>EXAONE-3.5-32B-Instruct</center></h1>"
+
+PLACEHOLDER = """
+<center>
+<p>Hi! How can I help you today?</p>
+</center>
+"""
 
-PLACEHOLDER = f'<h3><center>LongWriter-glm4-9b is trained based on glm-4-9b, and is capable of generating 10,000+ words at once.</center></h3>'
 
 CSS = """
 .duplicate-button {
-  margin: auto !important;
-  color: white !important;
-  background: black !important;
-  border-radius: 100vh !important;
+    margin: auto !important;
+    color: white !important;
+    background: black !important;
+    border-radius: 100vh !important;
+}
+h3 {
+    text-align: center;
 }
 """
 
+device = "cuda" # for GPU usage or "cpu" for CPU usage
+
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.bfloat16,
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_quant_type= "nf4")
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL)
 model = AutoModelForCausalLM.from_pretrained(
-        "THUDM/LongWriter-glm4-9b",
-        torch_dtype=torch.bfloat16,
-        device_map="auto",
-        trust_remote_code=True,
-        ).eval()
-
-tokenizer = AutoTokenizer.from_pretrained("THUDM/LongWriter-glm4-9b",trust_remote_code=True, use_fast=False)
-
-class StopOnTokens(StoppingCriteria):
-    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
-        # stop_ids = model.config.eos_token_id
-        stop_ids = [tokenizer.eos_token_id, tokenizer.get_command("<|user|>"),
-                    tokenizer.get_command("<|observation|>")]
-        for stop_id in stop_ids:
-            if input_ids[0][-1] == stop_id:
-                return True
-        return False
-
-@spaces.GPU()
-def stream_chat(message: str, history: list, temperature: float, max_new_tokens: int):
-    print(f'message is - {message}')
-    print(f'history is - {history}')
-    conversation = []
+    MODEL,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    quantization_config=quantization_config)
+
+@spaces.GPU(duration=100)
+def stream_chat(
+    message: str, 
+    history: list,
+    system_prompt: str,
+    temperature: float = 0.8, 
+    max_new_tokens: int = 1024, 
+    top_p: float = 1.0, 
+    top_k: int = 20, 
+    penalty: float = 1.2,
+):
+    print(f'message: {message}')
+    print(f'history: {history}')
+
+    conversation = [
+        {"role": "system", "content": system_prompt}
+    ]
     for prompt, answer in history:
-        conversation.extend([{"role": "user", "content": prompt}, {"role": "assistant", "content": answer}])
-    #conversation.append({"role": "user", "content": message})
+        conversation.extend([
+            {"role": "user", "content": prompt}, 
+            {"role": "assistant", "content": answer},
+        ])
 
-    print(f"Conversation is -\n{conversation}")
-    stop = StopOnTokens()
+    conversation.append({"role": "user", "content": message})
 
-    input_ids = tokenizer.build_chat_input(message, history=conversation, role='user').input_ids.to(next(model.parameters()).device)
-    #input_ids = tokenizer.apply_chat_template(conversation, tokenize=True, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device)
+    input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt").to(model.device)
+    
     streamer = TextIteratorStreamer(tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
-    eos_token_id = [tokenizer.eos_token_id, tokenizer.get_command("<|user|>"),
-                    tokenizer.get_command("<|observation|>")]
-
+    
     generate_kwargs = dict(
-        input_ids=input_ids,
+        input_ids=input_ids, 
+        max_new_tokens = max_new_tokens,
+        do_sample = False if temperature == 0 else True,
+        top_p = top_p,
+        top_k = top_k,
+        temperature = temperature,
+        repetition_penalty=penalty,
+        eos_token_id=255001,
         streamer=streamer,
-        max_new_tokens=max_new_tokens,
-        do_sample=True,
-        top_k=1,
-        temperature=temperature,
-        repetition_penalty=1,
-        stopping_criteria=StoppingCriteriaList([stop]),
-        eos_token_id=eos_token_id,
     )
-    #gen_kwargs = {**input_ids, **generate_kwargs}
-
-    thread = Thread(target=model.generate, kwargs=generate_kwargs)
-    thread.start()
-    buffer = ""   
-    for new_token in streamer:
-        if new_token and '<|user|>' not in new_token:
-            buffer += new_token
+
+    with torch.no_grad():
+        thread = Thread(target=model.generate, kwargs=generate_kwargs)
+        thread.start()
+        
+    buffer = ""
+    for new_text in streamer:
+        buffer += new_text
         yield buffer
 
-chatbot = gr.Chatbot(height=600, placeholder = PLACEHOLDER)
+            
+chatbot = gr.Chatbot(height=600, placeholder=PLACEHOLDER)
 
-with gr.Blocks(css=CSS) as demo:
+with gr.Blocks(css=CSS, theme="soft") as demo:
     gr.HTML(TITLE)
     gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button")
     gr.ChatInterface(
@@ -101,20 +107,52 @@ with gr.Blocks(css=CSS) as demo:
         fill_height=True,
         additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
         additional_inputs=[
+            gr.Textbox(
+                value="""
+                You are a helpful assistant.
+                """,
+                label="System Prompt",
+                lines=5,
+                render=False,
+            ),
             gr.Slider(
                 minimum=0,
                 maximum=1,
                 step=0.1,
-                value=0.5,
+                value=0.8,
                 label="Temperature",
                 render=False,
             ),
             gr.Slider(
-                minimum=1024,
-                maximum=32768,
+                minimum=128,
+                maximum=8192,
                 step=1,
-                value=4096,
-                label="Max New Tokens",
+                value=1024,
+                label="Max new tokens",
+                render=False,
+            ),
+            gr.Slider(
+                minimum=0.0,
+                maximum=1.0,
+                step=0.1,
+                value=1.0,
+                label="top_p",
+                render=False,
+            ),
+            gr.Slider(
+                minimum=1,
+                maximum=20,
+                step=1,
+                value=20,
+                label="top_k",
+                render=False,
+            ),
+            gr.Slider(
+                minimum=0.0,
+                maximum=2.0,
+                step=0.1,
+                value=1.2,
+                label="Repetition penalty",
                 render=False,
             ),
         ],
@@ -126,7 +164,7 @@ with gr.Blocks(css=CSS) as demo:
         ],
         cache_examples=False,
     )
-    
+
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()