first commit

app.py

@@ -1,62 +1,196 @@
+import os
+import time
+import spaces
+from threading import Thread
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 import gradio as gr
-from huggingface_hub import InferenceClient
 
+MODEL = "weblab-GENIAC/Tanuki-8B-dpo-v1.0"
+HF_TOKEN = os.environ.get("HF_TOKEN", None)
+
+TITLE = "<h1><center>Tanuki-8B-dpo-v1.0</center></h1>"
+
+DESCRIPTION = """
+<div class="model-description">
+    <p>
+        🦡 <a href="https://huggingface.co/weblab-GENIAC/Tanuki-8B-dpo-v1.0"><b>Tanuki 8B</b>(weblab-GENIAC/Tanuki-8B-dpo-v1.0)</a>は、
+        経産省及びNEDOが推進する日本国内の生成AI基盤モデル開発を推進する「GENIAC」プロジェクトにおいて、松尾・岩澤研究室が開発・公開したLLMとなります。
+        本プロジェクトは松尾研が提供する大規模言語モデル講座（2023年9月開催、2,000名が受講）の修了生及び一般公募によって集まった有志の開発者（⺠間企業・研究者・学⽣で構成）が、最新の研究成果や技術的な知見を取り入れ、開発を行ったモデルです。
+    </p>
+    <p>🤖 このデモでは、Tanuki 8Bとチャットを行うことが可能です。(注:フルバーションの<a href="https://huggingface.co/weblab-GENIAC/Tanuki-8x8B-dpo-v1.0">Tanuki 8x8B</a>ではございません。)</p>
+    <p>📄 モデルの詳細については、<a href="http://weblab.t.u-tokyo.ac.jp/2024-08-30">プレスリリース</a>をご覧ください。お問い合わせは<a href="https://weblab.t.u-tokyo.ac.jp/contact/">こちら</a>までどうぞ。</p>
+    <p>関連サイト: <a href="https://weblab.t.u-tokyo.ac.jp/geniac_llm">GENIAC 松尾研 LLM開発プロジェクト</a></p>
+</div>
 """
-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")
 
+PLACEHOLDER = """
+<div class="image-placeholder">
+    <img src="https://weblab.t.u-tokyo.ac.jp/wp-content/uploads/2024/06/GENIAC-image-cutting3-1.jpg" alt="Tanuki-8B Image">
+    <h1>Tanuki-8B</h1>
+</div>
+"""
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+CSS = """
+.duplicate-button {
+    margin: auto !important;
+    color: white !important;
+    background: black !important;
+    border-radius: 100vh !important;
+}
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+h3 {
+    text-align: center;
+}
 
-    messages.append({"role": "user", "content": message})
+.model-description {
+    padding: 0.5em 1em;
+    margin: 2em 0;
+    border-top: solid 5px #5d627b;
+    box-shadow: 0 1px 1px rgba(0, 0, 0, 0.22);
+    border-radius: 5px;
+}
 
-    response = ""
+.model-description p {
+    margin: 0;
+    padding: 0;
+    color: #5d627b;
+}
 
-    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
+.image-placeholder {
+    text-align: center;
+    display: flex;
+    flex-direction: column;
+    align-items: center;
+}
 
-        response += token
-        yield response
+.image-placeholder img {
+    width: 100%;
+    height: auto;
+    opacity: 0.55;
+}
 
+.image-placeholder h1 {
+    font-size: 28px;
+    margin-bottom: 2px;
+    opacity: 0.55;
+}
 """
-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)",
-        ),
-    ],
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL)
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
 )
+print(model)
+
+@spaces.GPU()
+def stream_chat(
+    message: str, 
+    history: list,
+    system_prompt: str,
+    temperature: float = 0.3, 
+    max_new_tokens: int = 1024, 
+    top_p: float = 1.0, 
+    top_k: int = 20, 
+):
+    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})
+
+    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)
+    
+    generate_kwargs = dict(
+        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,
+        streamer=streamer,
+    )
+
+    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)
+
+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.Markdown(DESCRIPTION)
+    gr.ChatInterface(
+        fn=stream_chat,
+        chatbot=chatbot,
+        fill_height=True,
+        additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
+        additional_inputs=[
+            gr.Textbox(
+                value="あなたは役に立つアシスタントです。",
+                label="System Prompt",
+                render=False,
+            ),
+            gr.Slider(
+                minimum=0,
+                maximum=1,
+                step=0.1,
+                value=0.3,
+                label="Temperature",
+                render=False,
+            ),
+            gr.Slider(
+                minimum=128,
+                maximum=8192,
+                step=1,
+                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,
+            ),
+        ],
+        examples=[
+            ["日本で有名なものと言えば"],
+            ["人工知能とは何ですか"],
+            ["C言語で素数を判定するコードを書いて"],
+            ["たぬきが主人公の物語を書いて"]
+        ],
+        cache_examples=False,
+    )
 
 
 if __name__ == "__main__":

requirements.txt

@@ -1 +1,6 @@
-huggingface_hub==0.22.2
+accelerate==0.33.0
+bitsandbytes==0.43.3
+torch==2.2.0
+transformers==4.44.0
+einops==0.8.0
+sentencepiece==0.2.0