Duplicate from matthoffner/falcon-40b-instruct-ggml

Co-authored-by: Matt Hoffner <matthoffner@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

Dockerfile

@@ -0,0 +1,66 @@
+FROM nvidia/cuda:12.0.0-cudnn8-devel-ubuntu22.04
+
+# Set environment variables
+ENV PATH="/usr/local/cuda/bin:$PATH"
+ENV MODEL_NAME="falcon-40b-instruct-GGML"
+ENV MODEL_FILE="falcon40b-instruct.ggmlv3.q4_K_S.bin"
+ENV MODEL_URL="https://huggingface.co/TheBloke/${MODEL_NAME}/raw/ggmlv3/${MODEL_FILE}"
+
+RUN apt update && \
+    apt install --no-install-recommends -y build-essential python3 python3-pip wget curl git && \
+    apt clean && rm -rf /var/lib/apt/lists/*
+
+# Set the working directory in the container to /app
+WORKDIR /app
+
+# Install cmake
+RUN apt-get install -y wget && \
+    wget -qO- "https://cmake.org/files/v3.18/cmake-3.18.0-Linux-x86_64.tar.gz" | tar --strip-components=1 -xz -C /usr/local
+
+# Copy the requirements.txt file into the container
+COPY requirements.txt ./
+
+# Install any needed packages specified in requirements.txt
+RUN pip3 install --upgrade pip && \
+    pip3 install -r requirements.txt
+
+# Clone ctransformers and update submodule ggllm.cpp
+RUN git clone --recursive https://github.com/marella/ctransformers.git && \
+    cd ctransformers && \
+    git submodule update --init models/submodules/ggllm.cpp && \
+    cd models/submodules/ggllm.cpp && \
+    git checkout master && \
+    git pull
+
+# Install ctransformers from source
+RUN cd ctransformers && \
+    CT_CUBLAS=1 FORCE_CMAKE=1 pip install .
+
+# Download the model file
+# RUN wget -O /app/${MODEL_FILE} ${MODEL_URL}
+
+# Create user
+RUN useradd -m -u 1000 user
+
+# Create a directory for app and move the downloaded file there
+# RUN mkdir -p /home/user/app && mv /app/${MODEL_FILE} /home/user/app
+
+# Change the ownership of the copied file to user
+# RUN chown user:user /home/user/app/${MODEL_FILE}
+
+USER user
+ENV HOME=/home/user \
+	PATH=/home/user/.local/bin:$PATH
+
+WORKDIR $HOME/app
+
+# Now you can COPY the rest of your app
+COPY --chown=user . .
+
+RUN ls -al
+
+# Make port available to the world outside this container
+EXPOSE 7860
+
+# Run uvicorn when the container launches
+CMD ["python3", "demo.py", "--host", "0.0.0.0", "--port", "7860"]

README.md

@@ -0,0 +1,15 @@
+---
+title: falcon-40b-ggml-cuda
+emoji: 🦅
+colorFrom: red
+colorTo: blue
+sdk: docker
+pinned: false
+app_port: 7860
+duplicated_from: matthoffner/falcon-40b-instruct-ggml
+---
+
+# falcon-40b-ggml-cuda
+
+## <a href="https://github.com/cmp-nct/ggllm.cpp" target="_blank">ggllm.cpp</a>
+## ctransformers

api.py

@@ -0,0 +1,79 @@
+import fastapi
+import json
+import uvicorn
+from fastapi import HTTPException
+from fastapi.responses import HTMLResponse
+from fastapi.middleware.cors import CORSMiddleware
+from sse_starlette.sse import EventSourceResponse
+from starlette.responses import StreamingResponse
+from ctransformers import AutoModelForCausalLM
+from pydantic import BaseModel
+from typing import List, Dict, Any, Generator
+
+llm = AutoModelForCausalLM.from_pretrained("TheBloke/falcon-40b-instruct-GGML", model_file="falcon40b-instruct.ggmlv3.q2_K.bin",
+                                           model_type="falcon", threads=8)
+app = fastapi.FastAPI(title="🦅Falcon 40B GGML (ggmlv3.q2_K)🦅")
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+class ChatCompletionRequestV0(BaseModel):
+    prompt: str
+
+class Message(BaseModel):
+    role: str
+    content: str
+
+class ChatCompletionRequest(BaseModel):
+    messages: List[Message]
+    max_tokens: int = 250
+
+@app.post("/v1/completions")
+async def completion(request: ChatCompletionRequestV0, response_mode=None):
+    response = llm(request.prompt)
+    return response
+
+@app.post("/v1/chat/completions")
+async def chat(request: ChatCompletionRequest):
+    combined_messages = ' '.join([message.content for message in request.messages])
+    tokens = llm.tokenize(combined_messages)
+    
+    try:
+        chat_chunks = llm.generate(tokens)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+    async def format_response(chat_chunks: Generator) -> Any:
+        for chat_chunk in chat_chunks:
+            response = {
+                'choices': [
+                    {
+                        'message': {
+                            'role': 'system',
+                            'content': llm.detokenize(chat_chunk)
+                        },
+                        'finish_reason': 'stop' if llm.detokenize(chat_chunk) == "[DONE]" else 'unknown'
+                    }
+                ]
+            }
+            yield f"data: {json.dumps(response)}\n\n"
+        yield "event: done\ndata: {}\n\n"
+
+    return StreamingResponse(format_response(chat_chunks), media_type="text/event-stream")
+
+@app.post("/v0/chat/completions")
+async def chat(request: ChatCompletionRequestV0, response_mode=None):
+    tokens = llm.tokenize(request.prompt)
+    async def server_sent_events(chat_chunks, llm):
+        for chat_chunk in llm.generate(chat_chunks):
+            yield dict(data=json.dumps(llm.detokenize(chat_chunk)))
+        yield dict(data="[DONE]")
+
+    return EventSourceResponse(server_sent_events(tokens, llm))
+
+if __name__ == "__main__":
+  uvicorn.run(app, host="0.0.0.0", port=8000)

demo.py

@@ -0,0 +1,198 @@
+import os
+from ctransformers import AutoModelForCausalLM
+import gradio as gr
+
+llm = AutoModelForCausalLM.from_pretrained("TheBloke/falcon-40b-instruct-GGML", model_file="falcon-40b-instruct.ggccv1.q2_k.bin",
+                                           model_type="falcon", threads=8, gpu_layers=30)
+
+TITLE = """<h2 align="center">🦅 falcon-40b-instruct.ggccv1.q2_k.bin 🦅"""
+USER_NAME = "User"
+BOT_NAME = "Falcon"
+DEFAULT_INSTRUCTIONS = f"""The following is a conversation between a highly knowledgeable and intelligent AI assistant, called Falcon, and a human user, called User. In the following interactions, User and Falcon will converse in natural language, and Falcon will answer User's questions. Falcon was built to be respectful, polite and inclusive. Falcon was built by the Technology Innovation Institute in Abu Dhabi. Falcon will never decline to answer a question, and always attempts to give an answer that User would be satisfied with. It knows a lot, and always tells the truth. The conversation begins.
+"""
+RETRY_COMMAND = "/retry"
+STOP_STR = f"\n{USER_NAME}:"
+STOP_SUSPECT_LIST = [":", "\n", "User"]
+
+
+def chat_accordion():
+    with gr.Accordion("Parameters", open=False):
+        temperature = gr.Slider(
+            minimum=0.1,
+            maximum=2.0,
+            value=0.8,
+            step=0.1,
+            interactive=True,
+            label="Temperature",
+        )
+        top_p = gr.Slider(
+            minimum=0.1,
+            maximum=0.99,
+            value=0.9,
+            step=0.01,
+            interactive=True,
+            label="p (nucleus sampling)",
+        )
+    return temperature, top_p
+
+
+def format_chat_prompt(message: str, chat_history, instructions: str) -> str:
+    instructions = instructions.strip(" ").strip("\n")
+    prompt = instructions
+    for turn in chat_history:
+        user_message, bot_message = turn
+        prompt = f"{prompt}\n{USER_NAME}: {user_message}\n{BOT_NAME}: {bot_message}"
+    prompt = f"{prompt}\n{USER_NAME}: {message}\n{BOT_NAME}:"
+    return prompt
+
+
+def chat():
+    with gr.Column(elem_id="chat_container"):
+        with gr.Row():
+            chatbot = gr.Chatbot(elem_id="chatbot")
+        with gr.Row():
+            inputs = gr.Textbox(
+                placeholder=f"Hello {BOT_NAME} !!",
+                label="Type an input and press Enter",
+                max_lines=3,
+            )
+
+    with gr.Row(elem_id="button_container"):
+        with gr.Column():
+            retry_button = gr.Button("♻️ Retry last turn")
+        with gr.Column():
+            delete_turn_button = gr.Button("🧽 Delete last turn")
+        with gr.Column():
+            clear_chat_button = gr.Button("✨ Delete all history")
+
+    gr.Examples(
+        [
+            ["Hey Falcon! Any recommendations for my holidays in Abu Dhabi?"],
+            ["What's the Everett interpretation of quantum mechanics?"],
+            ["Give me a list of the top 10 dive sites you would recommend around the world."],
+            ["Can you tell me more about deep-water soloing?"],
+            ["Can you write a short tweet about the Apache 2.0 release of our latest AI model, Falcon LLM?"],
+        ],
+        inputs=inputs,
+        label="Click on any example and press Enter in the input textbox!",
+    )
+
+    with gr.Row(elem_id="param_container"):
+        with gr.Column():
+            temperature, top_p = chat_accordion()
+        with gr.Column():
+            with gr.Accordion("Instructions", open=False):
+                instructions = gr.Textbox(
+                    placeholder="LLM instructions",
+                    value=DEFAULT_INSTRUCTIONS,
+                    lines=10,
+                    interactive=True,
+                    label="Instructions",
+                    max_lines=16,
+                    show_label=False,
+                )
+
+    def run_chat(message: str, chat_history, instructions: str, temperature: float, top_p: float):
+        if not message or (message == RETRY_COMMAND and len(chat_history) == 0):
+            yield chat_history
+            return
+
+        if message == RETRY_COMMAND and chat_history:
+            prev_turn = chat_history.pop(-1)
+            user_message, _ = prev_turn
+            message = user_message
+
+        prompt = format_chat_prompt(message, chat_history, instructions)
+        chat_history = chat_history + [[message, ""]]
+        stream = llm(
+            prompt,
+            max_new_tokens=1024,
+            stop=[STOP_STR, "<|endoftext|>"],
+            temperature=temperature,
+            top_p=top_p,
+            stream=True
+        )
+        acc_text = ""
+        for idx, response in enumerate(stream):
+            text_token = response
+
+            if text_token in STOP_SUSPECT_LIST:
+                acc_text += text_token
+                continue
+
+            if idx == 0 and text_token.startswith(" "):
+                text_token = text_token[1:]
+
+            acc_text += text_token
+            last_turn = list(chat_history.pop(-1))
+            last_turn[-1] += acc_text
+            chat_history = chat_history + [last_turn]
+            yield chat_history
+            acc_text = ""
+
+    def delete_last_turn(chat_history):
+        if chat_history:
+            chat_history.pop(-1)
+        return {chatbot: gr.update(value=chat_history)}
+
+    def run_retry(message: str, chat_history, instructions: str, temperature: float, top_p: float):
+        yield from run_chat(RETRY_COMMAND, chat_history, instructions, temperature, top_p)
+
+    def clear_chat():
+        return []
+
+    inputs.submit(
+        run_chat,
+        [inputs, chatbot, instructions, temperature, top_p],
+        outputs=[chatbot],
+        show_progress=False,
+    )
+    inputs.submit(lambda: "", inputs=None, outputs=inputs)
+    delete_turn_button.click(delete_last_turn, inputs=[chatbot], outputs=[chatbot])
+    retry_button.click(
+        run_retry,
+        [inputs, chatbot, instructions, temperature, top_p],
+        outputs=[chatbot],
+        show_progress=False,
+    )
+    clear_chat_button.click(clear_chat, [], chatbot)
+
+
+def get_demo():
+    with gr.Blocks(
+        # css=None
+        # css="""#chat_container {width: 700px; margin-left: auto; margin-right: auto;}
+        #        #button_container {width: 700px; margin-left: auto; margin-right: auto;}
+        #        #param_container {width: 700px; margin-left: auto; margin-right: auto;}"""
+        css="""#chatbot {
+    font-size: 14px;
+    min-height: 300px;
+}"""
+    ) as demo:
+        gr.HTML(TITLE)
+
+        with gr.Row():
+            with gr.Column():
+                gr.Markdown(
+                    """**Chat with [Falcon-40B-Instruct](https://huggingface.co/tiiuae/falcon-40b-instruct), brainstorm ideas, discuss your holiday plans, and more!**
+
+                    ✨ This demo is powered by [Falcon-40B](https://huggingface.co/tiiuae/falcon-40b), finetuned on the [Baize](https://github.com/project-baize/baize-chatbot) dataset, and running with [Text Generation Inference](https://github.com/huggingface/text-generation-inference). [Falcon-40B](https://huggingface.co/tiiuae/falcon-40b) is a state-of-the-art large language model built by the [Technology Innovation Institute](https://www.tii.ae) in Abu Dhabi. It is trained on 1 trillion tokens (including [RefinedWeb](https://huggingface.co/datasets/tiiuae/falcon-refinedweb)) and available under the Apache 2.0 license. It currently holds the 🥇 1st place on the [🤗 Open LLM leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard). This demo is made available by the [HuggingFace H4 team](https://huggingface.co/HuggingFaceH4). 
+
+                    🧪 This is only a **first experimental preview**: the [H4 team](https://huggingface.co/HuggingFaceH4) intends to provide increasingly capable versions of Falcon Chat in the future, based on improved datasets and RLHF/RLAIF.
+
+                    👀 **Learn more about Falcon LLM:** [falconllm.tii.ae](https://falconllm.tii.ae/)
+
+                    ➡️️ **Intended Use**: this demo is intended to showcase an early finetuning of [Falcon-40B](https://huggingface.co/tiiuae/falcon-40b), to illustrate the impact (and limitations) of finetuning on a dataset of conversations and instructions. We encourage the community to further build upon the base model, and to create even better instruct/chat versions!
+
+                    ⚠️ **Limitations**: the model can and will produce factually incorrect information, hallucinating facts and actions. As it has not undergone any advanced tuning/alignment, it can produce problematic outputs, especially if prompted to do so. Finally, this demo is limited to a session length of about 1,000 words.
+                    """
+                )
+
+        chat()
+
+    return demo
+
+if __name__ == "__main__":
+  demo = get_demo()
+  demo.queue(max_size=64, concurrency_count=8)
+  demo.launch(server_name="0.0.0.0", server_port=7860)

requirements.txt

@@ -0,0 +1,11 @@
+uvicorn
+markdown
+fastapi
+loguru
+torch
+numpy
+transformers
+accelerate
+langchain
+sse_starlette
+gradio