Upload 2 files

app.py

@@ -0,0 +1,198 @@
+import json
+import os
+import uuid
+from datetime import datetime
+from pathlib import Path
+from zoneinfo import ZoneInfo
+
+import gradio as gr
+from huggingface_hub import CommitScheduler
+from openai import OpenAI
+from transformers import AutoTokenizer
+
+openai_api_key = os.getenv("api_key")
+openai_api_base = os.getenv("api_url")
+hf_token = os.getenv("hf_token")
+model_name = "Aratako/calm3-22b-RP-v2"
+
+client = OpenAI(
+    api_key=openai_api_key,
+    base_url=openai_api_base,
+)
+
+tokenizer = AutoTokenizer.from_pretrained("Aratako/calm3-22b-RP-v2")
+
+# Define the file where to save the data. Use UUID to make sure not to overwrite existing data from a previous run.
+log_file = Path("logs/") / f"data_{uuid.uuid4()}.json"
+log_folder = log_file.parent
+
+# Schedule regular uploads. Remote repo and local folder are created if they don't already exist.
+scheduler = CommitScheduler(
+    repo_id="Aratako/calm3-22b-RP-v2-logs",  # Replace with your actual repo ID
+    repo_type="dataset",
+    folder_path=log_folder,
+    path_in_repo="data",
+    every=60,  # Upload every 60 minutes
+    token=hf_token,
+)
+
+
+def save_chat_logs(messages, response):
+    """
+    Save conversation data in a JSON Lines file.
+    """
+    with scheduler.lock:
+        entry = {
+            "timestamp": datetime.now(ZoneInfo("Asia/Tokyo")).isoformat(),
+            "messages": messages,
+            "response": response,
+        }
+        with log_file.open("a") as f:
+            f.write(json.dumps(entry, ensure_ascii=False) + "\n")
+
+
+def count_tokens(messages):
+    return sum(len(tokenizer.encode(msg["content"])) for msg in messages)
+
+
+def trim_messages(messages, max_tokens):
+    # システムメッセージは常に保持
+    system_message = messages[0]
+    trimmed_messages = [system_message]
+    current_tokens = count_tokens([system_message])
+
+    # 最新のメッセージから逆順に追加していく
+    for message in reversed(messages[1:]):
+        message_tokens = count_tokens([message])
+        if current_tokens + message_tokens <= max_tokens:
+            trimmed_messages.insert(1, message)  # システムメッセージの後に挿入
+            current_tokens += message_tokens
+        else:
+            break  # トークン制限を超えたら終了
+
+    return trimmed_messages
+
+
+def respond(
+    message,
+    history: list[tuple[str, str]],
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+):
+    try:
+        messages = [{"role": "system", "content": system_message}]
+
+        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})
+        # メッセージを調整して8192トークン以内に収める
+        max_input_tokens = 8192 - max_tokens
+        trimmed_messages = trim_messages(messages, max_input_tokens)
+
+        response = ""
+        for chunk in client.chat.completions.create(
+            model=model_name,
+            messages=trimmed_messages,
+            max_tokens=max_tokens,
+            stream=True,
+            temperature=temperature,
+            top_p=top_p,
+        ):
+            token = chunk.choices[0].delta.content
+            if token is not None:
+                response += token
+            yield response
+
+        # Save conversation after the full response is generated
+        save_chat_logs(trimmed_messages, response)
+
+    except Exception as e:
+        yield f"エラーが発生しました: {str(e)}"
+
+
+"""
+For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+"""
+
+description = """
+### [Aratako/calm3-22b-RP-v2](https://huggingface.co/Aratako/calm3-22b-RP-v2)のデモです。
+- 対話ログを収集します。ただしデータは非公開リポジトリに保存され、私（Aratako）以外の第三者に公開はしません。
+- 収集したログやそれを元に合成したデータを今後開発するモデルの学習に利用する可能性があります。ただしその際も本デモが元となるデータは公開しません。
+- 特定のシチュエーションや状況での対話を好む方は、積極的に使ってもらうと今後開発するモデルがその状況のRPが上手くなる可能性もあります。
+- **上記の条件に同意する場合のみ**､以下のChatbotを利用してください｡
+"""
+
+
+HEADER = description
+FOOTER = """### 注意
+- コンテクスト長は8192までです。超えた場合、古い対話から順番に削除されます。"""
+
+
+def run():
+    chatbot = gr.Chatbot(
+        elem_id="chatbot",
+        scale=1,
+        show_copy_button=True,
+        height="70%",
+        layout="panel",
+    )
+    with gr.Blocks(fill_height=True) as demo:
+        gr.Markdown(HEADER)
+        chat_interface = gr.ChatInterface(
+            fn=respond,
+            stop_btn="Stop Generation",
+            cache_examples=False,
+            multimodal=False,
+            chatbot=chatbot,
+            additional_inputs_accordion=gr.Accordion(
+                label="Parameters", open=False, render=False
+            ),
+            additional_inputs=[
+                gr.Textbox(
+                    value="ここにロールプレイの設定を書いてください。",
+                    label="システムメッセージ （ロールプレイの設定）",
+                    render=False,
+                ),
+                gr.Slider(
+                    minimum=1,
+                    maximum=4096,
+                    step=1,
+                    value=1024,
+                    label="Max tokens",
+                    visible=True,
+                    render=False,
+                ),
+                gr.Slider(
+                    minimum=0,
+                    maximum=1,
+                    step=0.1,
+                    value=0.7,
+                    label="Temperature",
+                    visible=True,
+                    render=False,
+                ),
+                gr.Slider(
+                    minimum=0,
+                    maximum=1,
+                    step=0.1,
+                    value=0.9,
+                    label="Top-p",
+                    visible=True,
+                    render=False,
+                ),
+            ],
+            analytics_enabled=False,
+        )
+        gr.Markdown(FOOTER)
+    demo.queue(max_size=256, api_open=True)
+    demo.launch(share=False, quiet=True)
+
+
+if __name__ == "__main__":
+    run()

requirements.txt

@@ -0,0 +1,3 @@
+huggingface_hub
+openai
+transformers