wip

README.md

@@ -1,19 +1,12 @@
 ---
-title: Code Llama 13B Chat
+title: Llama 7B Chat on Inf2
 emoji: 🦙
-colorFrom: indigo
-colorTo: pink
+colorFrom: yellow
+colorTo: purple
 sdk: gradio
-sdk_version: 3.37.0
+sdk_version: 3.46.1
 app_file: app.py
-pinned: false
+pinned: true
 license: other
-suggested_hardware: a10g-small
 duplicated_from: huggingface-projects/llama-2-13b-chat
 ---
-
-# LLAMA v2 Models
-
-Llama v2 was introduced in [this paper](https://arxiv.org/abs/2307.09288).
-
-This Space demonstrates [Llama-2-13b-chat-hf](meta-llama/Llama-2-13b-chat-hf) from Meta. Please, check the original model card for details.

app.py

@@ -1,279 +1,174 @@
 from typing import Iterator
-
 import gradio as gr
-import torch
-
-from model import get_input_token_length, run
-
-DEFAULT_SYSTEM_PROMPT = """\
-You are a helpful, respectful and honest assistant with a deep knowledge of code and software design. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.\n\nIf a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.\
-"""
-MAX_MAX_NEW_TOKENS = 4096
-DEFAULT_MAX_NEW_TOKENS = 1024
-MAX_INPUT_TOKEN_LENGTH = 4000
-
-DESCRIPTION = """
-# Code Llama 13B Chat
-
-This Space demonstrates model [CodeLlama-13b-Instruct](https://huggingface.co/codellama/CodeLlama-13b-Instruct-hf) by Meta, a Code Llama model with 13B parameters fine-tuned for chat instructions and specialized on code tasks. Feel free to play with it, or duplicate to run generations without a queue! If you want to run your own service, you can also [deploy the model on Inference Endpoints](https://huggingface.co/inference-endpoints).
-
-🔎 For more details about the Code Llama family of models and how to use them with `transformers`, take a look [at our blog post](https://huggingface.co/blog/codellama) or [the paper](https://huggingface.co/papers/2308.12950).
-
-🏃🏻 Check out our [Playground](https://huggingface.co/spaces/codellama/codellama-playground) for a super-fast code completion demo that leverages a streaming [inference endpoint](https://huggingface.co/inference-endpoints).
-
-💪 For a chat demo of the largest Code Llama model (34B parameters), you can [select Code Llama in Hugging Chat!](https://huggingface.co/chat/)
-
-"""
-
-LICENSE = """
-<p/>
-
----
-As a derivate work of Code Llama by Meta,
-this demo is governed by the original [license](https://huggingface.co/spaces/huggingface-projects/codellama-2-13b-chat/blob/main/LICENSE.txt) and [acceptable use policy](https://huggingface.co/spaces/huggingface-projects/codellama-2-13b-chat/blob/main/USE_POLICY.md).
-"""
-
-if not torch.cuda.is_available():
-    DESCRIPTION += '\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>'
-
-
-def clear_and_save_textbox(message: str) -> tuple[str, str]:
-    return '', message
-
-
-def display_input(message: str,
-                  history: list[tuple[str, str]]) -> list[tuple[str, str]]:
-    history.append((message, ''))
-    return history
-
-
-def delete_prev_fn(
-        history: list[tuple[str, str]]) -> tuple[list[tuple[str, str]], str]:
-    try:
-        message, _ = history.pop()
-    except IndexError:
-        message = ''
-    return history, message or ''
+import boto3
+import io
+import json
+import os
+from transformers import AutoTokenizer
+
+aws_access_key_id = os.environ.get("AWS_ACCESS_KEY_ID", None)
+aws_secret_access_key = os.environ.get("AWS_SECRET_ACCESS_KEY", None)
+aws_session_token = os.environ.get("AWS_SESSION_TOKEN", None)
+region = os.environ.get("AWS_REGION", None)
+endpoint_name = os.environ.get("SAGEMAKER_ENDPOINT_NAME", None)
+
+tokenizer = AutoTokenizer.from_pretrained(
+    "aws-neuron/Llama-2-7b-chat-hf-seqlen-2048-bs-4"
+)
+
+# if (
+#     aws_access_key_id is None
+#     or aws_secret_access_key is None
+#     or region is None
+#     or endpoint_name is None
+# ):
+#     raise Exception(
+#         "Please set AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION and SAGEMAKER_ENDPOINT_NAME environment variables"
+#     )
+
+# boto_session = boto3.Session(
+#     aws_access_key_id=aws_access_key_id,
+#     aws_secret_access_key=aws_secret_access_key,
+#     aws_session_token=aws_session_token,
+#     region_name=region,
+# )
+
+# smr = boto_session.client("sagemaker-runtime")
+
+
+DEFAULT_SYSTEM_PROMPT = (
+    "You are an helpful Assistant, called Llama. Knowing everyting about AWS."
+)
+MAX_INPUT_TOKEN_LENGTH = 1024
+
+# hyperparameters for llm
+parameters = {
+    "do_sample": True,
+    "top_p": 0.9,
+    "temperature": 0.8,
+    "max_new_tokens": 1024,
+    "repetition_penalty": 1.03,
+    "stop": ["<\s>"],
+}
+
+
+# Helper for reading lines from a stream
+class LineIterator:
+    def __init__(self, stream):
+        self.byte_iterator = iter(stream)
+        self.buffer = io.BytesIO()
+        self.read_pos = 0
+
+    def __iter__(self):
+        return self
+
+    def __next__(self):
+        while True:
+            self.buffer.seek(self.read_pos)
+            line = self.buffer.readline()
+            if line and line[-1] == ord("\n"):
+                self.read_pos += len(line)
+                return line[:-1]
+            try:
+                chunk = next(self.byte_iterator)
+            except StopIteration:
+                if self.read_pos < self.buffer.getbuffer().nbytes:
+                    continue
+                raise
+            if "PayloadPart" not in chunk:
+                print("Unknown event type:" + chunk)
+                continue
+            self.buffer.seek(0, io.SEEK_END)
+            self.buffer.write(chunk["PayloadPart"]["Bytes"])
+
+
+def format_prompt(message, history):
+    messages = [{"role": "system", "content": DEFAULT_SYSTEM_PROMPT}]
+    for interaction in history:
+        messages.append({"role": "user", "content": interaction[0]})
+        messages.append({"role": "assistant", "content": interaction[1]})
+    messages.append({"role": "user", "content": message})
+    prompt = tokenizer.apply_chat_template(
+        messages, tokenize=False, add_generation_prompt=True
+    )
+    return prompt
 
 
 def generate(
-    message: str,
-    history_with_input: list[tuple[str, str]],
-    system_prompt: str,
-    max_new_tokens: int,
-    temperature: float,
-    top_p: float,
-    top_k: int,
-) -> Iterator[list[tuple[str, str]]]:
-    if max_new_tokens > MAX_MAX_NEW_TOKENS:
-        raise ValueError
-
-    history = history_with_input[:-1]
-    generator = run(message, history, system_prompt, max_new_tokens, temperature, top_p, top_k)
-    try:
-        first_response = next(generator)
-        yield history + [(message, first_response)]
-    except StopIteration:
-        yield history + [(message, '')]
-    for response in generator:
-        yield history + [(message, response)]
-
-
-def process_example(message: str) -> tuple[str, list[tuple[str, str]]]:
-    generator = generate(message, [], DEFAULT_SYSTEM_PROMPT, 1024, 1, 0.95, 50)
-    for x in generator:
-        pass
-    return '', x
-
-
-def check_input_token_length(message: str, chat_history: list[tuple[str, str]], system_prompt: str) -> None:
-    input_token_length = get_input_token_length(message, chat_history, system_prompt)
+    prompt,
+    history,
+):
+    formatted_prompt = format_prompt(prompt, history)
+    check_input_token_length(formatted_prompt)
+
+    request = {"inputs": formatted_prompt, "parameters": parameters, "stream": True}
+    resp = {"Body": ""}
+    # resp = {"Body": open("test.json", "rb")}
+    # resp = smr.invoke_endpoint_with_response_stream(
+    #     EndpointName=endpoint_name,
+    #     Body=json.dumps(request),
+    #     ContentType="application/json",
+    # )
+
+    output = "offline"
+    # for c in LineIterator(resp["Body"]):
+    #     c = c.decode("utf-8")
+    #     if c.startswith("data:"):
+    #         chunk = json.loads(c.lstrip("data:").rstrip("/n"))
+    #         if chunk["token"]["special"]:
+    #             continue
+    #         if chunk["token"]["text"] in request["parameters"]["stop"]:
+    #             break
+    #         output += chunk["token"]["text"]
+    #         for stop_str in request["parameters"]["stop"]:
+    #             if output.endswith(stop_str):
+    #                 output = output[: -len(stop_str)]
+    #                 output = output.rstrip()
+    #                 yield output
+
+    #         yield output
+    return output
+
+
+def check_input_token_length(prompt: str) -> None:
+    input_token_length = len(tokenizer(prompt)["input_ids"])
     if input_token_length > MAX_INPUT_TOKEN_LENGTH:
-        raise gr.Error(f'The accumulated input is too long ({input_token_length} > {MAX_INPUT_TOKEN_LENGTH}). Clear your chat history and try again.')
-
-
-with gr.Blocks(css='style.css') as demo:
-    gr.Markdown(DESCRIPTION)
-    gr.DuplicateButton(value='Duplicate Space for private use',
-                       elem_id='duplicate-button')
-
-    with gr.Group():
-        chatbot = gr.Chatbot(label='Chatbot')
-        with gr.Row():
-            textbox = gr.Textbox(
-                container=False,
-                show_label=False,
-                placeholder='Type a message...',
-                scale=10,
-            )
-            submit_button = gr.Button('Submit',
-                                      variant='primary',
-                                      scale=1,
-                                      min_width=0)
-    with gr.Row():
-        retry_button = gr.Button('🔄  Retry', variant='secondary')
-        undo_button = gr.Button('↩️ Undo', variant='secondary')
-        clear_button = gr.Button('🗑️  Clear', variant='secondary')
-
-    saved_input = gr.State()
-
-    with gr.Accordion(label='Advanced options', open=False):
-        system_prompt = gr.Textbox(label='System prompt',
-                                   value=DEFAULT_SYSTEM_PROMPT,
-                                   lines=6)
-        max_new_tokens = gr.Slider(
-            label='Max new tokens',
-            minimum=1,
-            maximum=MAX_MAX_NEW_TOKENS,
-            step=1,
-            value=DEFAULT_MAX_NEW_TOKENS,
-        )
-        temperature = gr.Slider(
-            label='Temperature',
-            minimum=0.1,
-            maximum=4.0,
-            step=0.1,
-            value=0.1,
-        )
-        top_p = gr.Slider(
-            label='Top-p (nucleus sampling)',
-            minimum=0.05,
-            maximum=1.0,
-            step=0.05,
-            value=0.9,
-        )
-        top_k = gr.Slider(
-            label='Top-k',
-            minimum=1,
-            maximum=1000,
-            step=1,
-            value=10,
+        raise gr.Error(
+            f"The accumulated input is too long ({input_token_length} > {MAX_INPUT_TOKEN_LENGTH}). Clear your chat history and try again."
         )
 
-    gr.Examples(
-        examples=[
-            'What is the Fibonacci sequence?',
-            'Can you explain briefly what Python is good for?',
-            'How can I display a grid of images in SwiftUI?',
-        ],
-        inputs=textbox,
-        outputs=[textbox, chatbot],
-        fn=process_example,
-        cache_examples=True,
-    )
-
-    gr.Markdown(LICENSE)
-
-    textbox.submit(
-        fn=clear_and_save_textbox,
-        inputs=textbox,
-        outputs=[textbox, saved_input],
-        api_name=False,
-        queue=False,
-    ).then(
-        fn=display_input,
-        inputs=[saved_input, chatbot],
-        outputs=chatbot,
-        api_name=False,
-        queue=False,
-    ).then(
-        fn=check_input_token_length,
-        inputs=[saved_input, chatbot, system_prompt],
-        api_name=False,
-        queue=False,
-    ).success(
-        fn=generate,
-        inputs=[
-            saved_input,
-            chatbot,
-            system_prompt,
-            max_new_tokens,
-            temperature,
-            top_p,
-            top_k,
-        ],
-        outputs=chatbot,
-        api_name=False,
-    )
-
-    button_event_preprocess = submit_button.click(
-        fn=clear_and_save_textbox,
-        inputs=textbox,
-        outputs=[textbox, saved_input],
-        api_name=False,
-        queue=False,
-    ).then(
-        fn=display_input,
-        inputs=[saved_input, chatbot],
-        outputs=chatbot,
-        api_name=False,
-        queue=False,
-    ).then(
-        fn=check_input_token_length,
-        inputs=[saved_input, chatbot, system_prompt],
-        api_name=False,
-        queue=False,
-    ).success(
-        fn=generate,
-        inputs=[
-            saved_input,
-            chatbot,
-            system_prompt,
-            max_new_tokens,
-            temperature,
-            top_p,
-            top_k,
-        ],
-        outputs=chatbot,
-        api_name=False,
-    )
 
-    retry_button.click(
-        fn=delete_prev_fn,
-        inputs=chatbot,
-        outputs=[chatbot, saved_input],
-        api_name=False,
-        queue=False,
-    ).then(
-        fn=display_input,
-        inputs=[saved_input, chatbot],
-        outputs=chatbot,
-        api_name=False,
-        queue=False,
-    ).then(
-        fn=generate,
-        inputs=[
-            saved_input,
-            chatbot,
-            system_prompt,
-            max_new_tokens,
-            temperature,
-            top_p,
-            top_k,
-        ],
-        outputs=chatbot,
-        api_name=False,
-    )
+theme = gr.themes.Monochrome(
+    primary_hue="indigo",
+    secondary_hue="blue",
+    neutral_hue="slate",
+    radius_size=gr.themes.sizes.radius_sm,
+    font=[
+        gr.themes.GoogleFont("Open Sans"),
+        "ui-sans-serif",
+        "system-ui",
+        "sans-serif",
+    ],
+)
+DESCRIPTION = """
+<div style="text-align: center; max-width: 650px; margin: 0 auto; display:grid; gap:25px;">
+    <img class="logo" src="https://huggingface.co/datasets/philschmid/assets/resolve/main/aws-neuron_hf.png" alt="Hugging Face Neuron Logo"
+        style="margin: auto; max-width: 14rem;">        
+    <h1 style="font-weight: 900; margin-bottom: 7px;margin-top:5px">
+        Llama 2 7B Chat on AWS INF2 ⚡
+    </h1> 
+    <p style="margin-bottom: 10px; font-size: 94%; line-height: 23px;">
+    Llama 2 is a collection of pretrained and fine-tuned generative text models ranging in scale from 7 billion to 70 billion parameters. This is the repository for the 7B fine-tuned model, optimized for dialogue use cases and converted for the Hugging Face Transformers format. Links to other models can be found in the index at the bottom. This demo is running on <a style="text-decoration: underline;" href="https://aws.amazon.com/ec2/instance-types/inf2/?nc1=h_ls">AWS Inferentia2</a>, <a href="https://www.philschmid.de/inferentia2-llama-7b" target="_blank">How does it work?</a>
+    </p>
+</div>
+"""
 
-    undo_button.click(
-        fn=delete_prev_fn,
-        inputs=chatbot,
-        outputs=[chatbot, saved_input],
-        api_name=False,
-        queue=False,
-    ).then(
-        fn=lambda x: x,
-        inputs=[saved_input],
-        outputs=textbox,
-        api_name=False,
-        queue=False,
-    )
 
-    clear_button.click(
-        fn=lambda: ([], ''),
-        outputs=[chatbot, saved_input],
-        queue=False,
-        api_name=False,
-    )
+demo = gr.ChatInterface(
+    generate,
+    description=DESCRIPTION,
+    chatbot=gr.Chatbot(layout="panel"),
+    theme=theme,
+)
 
-demo.queue(max_size=20).launch()
+demo.queue(concurrency_count=5).launch(share=False)

model.py

@@ -1,75 +0,0 @@
-from threading import Thread
-from typing import Iterator
-
-import torch
-from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
-
-model_id = 'codellama/CodeLlama-13b-Instruct-hf'
-
-if torch.cuda.is_available():
-    config = AutoConfig.from_pretrained(model_id)
-    config.pretraining_tp = 1
-    model = AutoModelForCausalLM.from_pretrained(
-        model_id,
-        config=config,
-        torch_dtype=torch.float16,
-        load_in_4bit=True,
-        device_map='auto',
-        use_safetensors=False,
-    )
-else:
-    model = None
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-
-
-def get_prompt(message: str, chat_history: list[tuple[str, str]],
-               system_prompt: str) -> str:
-    texts = [f'<s>[INST] <<SYS>>\n{system_prompt}\n<</SYS>>\n\n']
-    # The first user input is _not_ stripped
-    do_strip = False
-    for user_input, response in chat_history:
-        user_input = user_input.strip() if do_strip else user_input
-        do_strip = True
-        texts.append(f'{user_input} [/INST] {response.strip()} </s><s>[INST] ')
-    message = message.strip() if do_strip else message
-    texts.append(f'{message} [/INST]')
-    return ''.join(texts)
-
-
-def get_input_token_length(message: str, chat_history: list[tuple[str, str]], system_prompt: str) -> int:
-    prompt = get_prompt(message, chat_history, system_prompt)
-    input_ids = tokenizer([prompt], return_tensors='np', add_special_tokens=False)['input_ids']
-    return input_ids.shape[-1]
-
-
-def run(message: str,
-        chat_history: list[tuple[str, str]],
-        system_prompt: str,
-        max_new_tokens: int = 1024,
-        temperature: float = 0.1,
-        top_p: float = 0.9,
-        top_k: int = 50) -> Iterator[str]:
-    prompt = get_prompt(message, chat_history, system_prompt)
-    inputs = tokenizer([prompt], return_tensors='pt', add_special_tokens=False).to('cuda')
-
-    streamer = TextIteratorStreamer(tokenizer,
-                                    timeout=10.,
-                                    skip_prompt=True,
-                                    skip_special_tokens=True)
-    generate_kwargs = dict(
-        inputs,
-        streamer=streamer,
-        max_new_tokens=max_new_tokens,
-        do_sample=True,
-        top_p=top_p,
-        top_k=top_k,
-        temperature=temperature,
-        num_beams=1,
-    )
-    t = Thread(target=model.generate, kwargs=generate_kwargs)
-    t.start()
-
-    outputs = []
-    for text in streamer:
-        outputs.append(text)
-        yield ''.join(outputs)

requirements.txt

@@ -1,8 +1,3 @@
-accelerate
-bitsandbytes
-gradio
-protobuf
-scipy
-sentencepiece
-torch
-git+https://github.com/huggingface/transformers@main
+boto3
+sagemaker
+transformers