feat: tanuki 8b instruct

app.py

@@ -1,12 +1,30 @@
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
+from threading import Thread
+
 import gradio as gr
-from huggingface_hub import InferenceClient
 
-"""
-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")
+try:
+    import spaces
+except:
+
+    class spaces:
+        @staticmethod
+        def GPU(duration: int):
+            return lambda x: x
+
+
+MODEL_NAME = "hatakeyama-llm-team/Tanuki-8B-Instruct"
+
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_NAME, load_in_8bit=True, device_map="auto"
+)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+
+print(model.hf_device_map)
 
 
+@spaces.GPU(duration=10)
 def respond(
     message,
     history: list[tuple[str, str]],
@@ -14,6 +32,7 @@ def respond(
     max_tokens,
     temperature,
     top_p,
+    top_k,
 ):
     messages = [{"role": "system", "content": system_message}]
 
@@ -25,27 +44,39 @@ def respond(
 
     messages.append({"role": "user", "content": message})
 
-    response = ""
+    tokenized_input = tokenizer.apply_chat_template(
+        messages, add_generation_prompt=True, tokenize=True, return_tensors="pt"
+    ).to(model.device)
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
+    streamer = TextIteratorStreamer(
+        tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True
+    )
+    generate_kwargs = dict(
+        input_ids=tokenized_input,
+        streamer=streamer,
+        max_new_tokens=max_tokens,
+        do_sample=True,
         temperature=temperature,
+        top_k=top_k,
         top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+        num_beams=1,
+    )
+    t = Thread(target=model.generate, kwargs=generate_kwargs)
+    t.start()
+
+    partial_message = ""
+    for new_token in streamer:
+        partial_message += new_token
+        yield partial_message
 
-        response += token
-        yield response
 
-"""
-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.Textbox(
+            value="以下は、タスクを説明する指示と、文脈のある入力の組み合わせです。要求を適切に満たす応答を書きなさい。",
+            label="システムプロンプト",
+        ),
         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(
@@ -53,11 +84,18 @@ demo = gr.ChatInterface(
             maximum=1.0,
             value=0.95,
             step=0.05,
-            label="Top-p (nucleus sampling)",
+            label="Top-p",
         ),
+        gr.Slider(minimum=1, maximum=2000, value=200, step=10, label="Top-k"),
+    ],
+    examples=[
+        ["たぬきってなんですか？"],
+        ["情けは人の為ならずとはどういう意味ですか？"],
+        ["明晰夢とはなんですか？"],
+        ["シュレディンガー方程式とシュレディンガーの猫はどのような関係がありますか？"],
     ],
 )
 
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()

requirements.txt

@@ -1 +1,5 @@
-huggingface_hub==0.22.2
+torch==2.3.0
+accelerate==0.30.1
+transformers==4.41.2
+optimum-quanto==0.2.1
+spaces==0.28.3