init modified demo

README.md

@@ -1,13 +1,46 @@
 ---
-title: Beecoder Playground
-emoji: 🐠
-colorFrom: purple
-colorTo: blue
+title: BeeCoder Demo
+emoji: 🐝
+colorFrom: gray
+colorTo: yellow
 sdk: gradio
-sdk_version: 4.0.2
+sdk_version: 3.28.3
 app_file: app.py
-pinned: false
+pinned: true
 license: apache-2.0
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# 🐝BeeCoder Demo🐝
+
+## Code-Completion Playground 💻 with 🐝[BeeCoder](https://huggingface.co/BEE-spoke-data/smol_llama-101M-GQA-python) Models
+
+This is a demo playground for generating Python code with the power of 🐝[BeeCoder](https://huggingface.co/BEE-spoke-data/smol_llama-101M-GQA-python), a **fine-tuned** version of the tiny [101M base model](https://huggingface.co/BEE-spoke-data/smol_llama-101M-GQA) on a dataset of pypi packages.
+
+ℹ️ This is not an instruction model but just a code completion tool.
+
+---
+
+**Intended Use**: This app and its [supporting model](https://huggingface.co/BEE-spoke-data/smol_llama-101M-GQA-python) are provided for demonstration purposes only; not to serve as a replacement for human expertise. For more details on the model, please refer to the [model card](https://huggingface.co/BEE-spoke-data/smol_llama-101M-GQA-python).
+
+In our country, we say _"To let 100M parameters model generate python script and not validate is like to let monkey fly a plane"_. So please be careful with the generated code.
+
+---
+
+## Base Model Information
+
+The base model, smol_llama-101M-GQA, was pretrained on a relatively few (< ~20B) high-quality tokens. It is tiny in size (101M parameters) but relatively powerful in performance. The training for the base model included datasets such as:
+
+- [JeanKaddour/minipile](https://huggingface.co/datasets/JeanKaddour/minipile)
+- [pszemraj/simple_wikipedia_LM](https://huggingface.co/datasets/pszemraj/simple_wikipedia_LM)
+- [BEE-spoke-data/wikipedia-20230901.en-deduped](https://huggingface.co/datasets/BEE-spoke-data/wikipedia-20230901.en-deduped)
+- [mattymchen/refinedweb-3m](https://huggingface.co/datasets/mattymchen/refinedweb-3m)
+
+You can find more information about the base model [here](https://huggingface.co/BEE-spoke-data/smol_llama-101M-GQA).
+
+---
+
+### Credits
+
+This app is modified from a demo playground originally built for [StarCoder](https://huggingface.co/bigcode/starcoder) by [BigCode](https://huggingface.co/bigcode).  You can find the original demo [here](https://huggingface.co/spaces/bigcode/bigcode-playground).
+
+---

app.py

@@ -0,0 +1,242 @@
+import gradio as gr
+import torch
+from gradio.themes.utils import sizes
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+import utils
+from constants import END_OF_TEXT
+from settings import DEFAULT_PORT
+
+# Load the tokenizer and model
+tokenizer = AutoTokenizer.from_pretrained(
+    "BEE-spoke-data/smol_llama-101M-GQA-python",
+    use_fast=False,
+)
+tokenizer.pad_token_id = tokenizer.eos_token_id
+tokenizer.pad_token = END_OF_TEXT
+model = AutoModelForCausalLM.from_pretrained(
+    "BEE-spoke-data/smol_llama-101M-GQA-python",
+    device_map="auto",
+)
+model = torch.compile(model, mode="reduce-overhead")
+
+# UI things
+
+_styles = utils.get_file_as_string("styles.css")
+
+# Loads ./README.md file & splits it into sections
+readme_file_content = utils.get_file_as_string("README.md", path="./")
+(
+    manifest,
+    description,
+    disclaimer,
+    base_model_info,
+    formats,
+) = utils.get_sections(readme_file_content, "---", up_to=5)
+
+theme = gr.themes.Soft(
+    primary_hue="yellow",
+    secondary_hue="orange",
+    neutral_hue="slate",
+    radius_size=sizes.radius_sm,
+    font=[
+        gr.themes.GoogleFont("IBM Plex Sans", [400, 600]),
+        "ui-sans-serif",
+        "system-ui",
+        "sans-serif",
+    ],
+    text_size=sizes.text_lg,
+)
+
+
+def run_inference(prompt, temperature, max_new_tokens, top_p, repetition_penalty):
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+    outputs = model.generate(
+        **inputs,
+        max_new_tokens=max_new_tokens,
+        min_new_tokens=8,
+        renormalize_logits=True,
+        no_repeat_ngram_size=6,
+        repetition_penalty=repetition_penalty,
+        num_beams=3,
+        early_stopping=True,
+        do_sample=True,
+        temperature=temperature,
+        top_p=top_p,
+    )
+    text = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
+    return text
+
+
+# Gradio interface wrapper for inference
+def gradio_interface(
+    prompt: str,
+    temperature: float,
+    max_new_tokens: int,
+    top_p: float,
+    repetition_penalty: float,
+):
+    return run_inference(prompt, temperature, max_new_tokens, top_p, repetition_penalty)
+
+
+import random
+
+examples = [
+    ["def add_numbers(a, b):\n    return", 0.2, 192, 0.9, 1.2],
+    [
+        "class Car:\n    def __init__(self, make, model):\n        self.make = make\n        self.model = model\n\n    def display_car(self):",
+        0.2,
+        192,
+        0.9,
+        1.2,
+    ],
+    [
+        "import pandas as pd\ndata = {'Name': ['Tom', 'Nick', 'John'], 'Age': [20, 21, 19]}\ndf = pd.DataFrame(data).convert_dtypes()\n# eda",
+        0.2,
+        192,
+        0.9,
+        1.2,
+    ],
+    [
+        "def factorial(n):\n    if n == 0:\n        return 1\n    else:",
+        0.2,
+        192,
+        0.9,
+        1.2,
+    ],
+    [
+        'def fibonacci(n):\n    if n <= 0:\n        raise ValueError("Incorrect input")\n    elif n == 1:\n        return 0\n    elif n == 2:\n        return 1\n    else:',
+        0.2,
+        192,
+        0.9,
+        1.2,
+    ],
+    [
+        "import matplotlib.pyplot as plt\nimport numpy as np\nx = np.linspace(0, 10, 100)\n# simple plot",
+        0.2,
+        192,
+        0.9,
+        1.2,
+    ],
+    ["def reverse_string(s:str) -> str:\n    return", 0.2, 192, 0.9, 1.2],
+    ["def is_palindrome(word:str) -> bool:\n    return", 0.2, 192, 0.9, 1.2],
+    [
+        "def bubble_sort(lst: list):\n    n = len(lst)\n    for i in range(n):\n        for j in range(0, n-i-1):",
+        0.2,
+        192,
+        0.9,
+        1.2,
+    ],
+    [
+        "def binary_search(arr, low, high, x):\n    if high >= low:\n        mid = (high + low) // 2\n        if arr[mid] == x:\n            return mid\n        elif arr[mid] > x:",
+        0.2,
+        192,
+        0.9,
+        1.2,
+    ],
+]
+
+# Define the Gradio Blocks interface
+with gr.Blocks(theme=theme, analytics_enabled=False, css=_styles) as demo:
+    with gr.Column():
+        gr.Markdown(description)
+        with gr.Row():
+            with gr.Column():
+                instruction = gr.Textbox(
+                    value=random.choice([e[0] for e in examples]),
+                    placeholder="Enter your code here",
+                    label="Code",
+                    elem_id="q-input",
+                )
+                submit = gr.Button("Generate", variant="primary")
+                output = gr.Code(elem_id="q-output", language="python", lines=10)
+                with gr.Row():
+                    with gr.Column():
+                        with gr.Accordion("Advanced settings", open=False):
+                            with gr.Row():
+                                column_1, column_2 = gr.Column(), gr.Column()
+                                with column_1:
+                                    temperature = gr.Slider(
+                                        label="Temperature",
+                                        value=0.2,
+                                        minimum=0.0,
+                                        maximum=1.0,
+                                        step=0.05,
+                                        interactive=True,
+                                        info="Higher values produce more diverse outputs",
+                                    )
+                                    max_new_tokens = gr.Slider(
+                                        label="Max new tokens",
+                                        value=128,
+                                        minimum=0,
+                                        maximum=512,
+                                        step=64,
+                                        interactive=True,
+                                        info="Number of tokens to generate",
+                                    )
+                                with column_2:
+                                    top_p = gr.Slider(
+                                        label="Top-p (nucleus sampling)",
+                                        value=0.90,
+                                        minimum=0.0,
+                                        maximum=1,
+                                        step=0.05,
+                                        interactive=True,
+                                        info="Higher values sample more low-probability tokens",
+                                    )
+                                    repetition_penalty = gr.Slider(
+                                        label="Repetition penalty",
+                                        value=1.1,
+                                        minimum=1.0,
+                                        maximum=2.0,
+                                        step=0.05,
+                                        interactive=True,
+                                        info="Penalize repeated tokens",
+                                    )
+                    with gr.Column():
+                        version = gr.Dropdown(
+                            [
+                                "smol_llama-101M-GQA-python",
+                            ],
+                            value="smol_llama-101M-GQA-python",
+                            label="Version",
+                            info="",
+                        )
+                gr.Markdown(disclaimer)
+                gr.Examples(
+                    examples=examples,
+                    inputs=[
+                        instruction,
+                        temperature,
+                        max_new_tokens,
+                        top_p,
+                        repetition_penalty,
+                        version,
+                    ],
+                    cache_examples=False,
+                    fn=gradio_interface,
+                    outputs=[output],
+                )
+                gr.Markdown(base_model_info)
+                gr.Markdown(formats)
+
+    submit.click(
+        gradio_interface,
+        inputs=[
+            instruction,
+            temperature,
+            max_new_tokens,
+            top_p,
+            repetition_penalty,
+        ],
+        outputs=[output],
+        # preprocess=False,
+        max_batch_size=2,
+        show_progress=True,
+    )
+
+demo.queue(max_size=10).launch(
+    debug=True,
+    server_port=DEFAULT_PORT,
+    max_threads=utils.get_workers(),
+)

constants.py

@@ -0,0 +1,4 @@
+END_OF_TEXT = "<|endoftext|>"
+
+# Near zero temperature to avoid division by zero
+MIN_TEMPERATURE = 1e-4

settings.py

@@ -0,0 +1,16 @@
+# URLs for the StarCoder Models/APIs
+DEFAULT_HUGGINGFACE_MODELS_API_BASE_URL = "https://api-inference.huggingface.co/models/"
+DEFAULT_STARCODER_API_PATH = "bigcode/starcoder/"
+DEFAULT_STARCODER_BASE_API_PATH = "bigcode/starcoderbase/"
+FIM_INDICATOR = "<FILL_HERE>"
+DEFAULT_PORT = 7860
+
+STATIC_PATH = "static"
+
+DEFAULT_SETTINGS = dict(
+    temperature=0.9,
+    max_new_tokens=256,
+    top_p=0.95,
+    repetition_penalty=1.0,
+    version="StarCoder",
+)

static/styles.css

@@ -0,0 +1,78 @@
+@import url('https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;600;700&display=swap');
+
+h1, h2 {
+    font-family: 'IBM Plex Mono', sans-serif;
+}
+
+.generating {
+    visibility: hidden
+}
+
+.gradio-container {
+    color: black
+}
+
+/* monospace_css */
+#q-input textarea {
+    font-family: monospace, 'Consolas', Courier, monospace;
+}
+
+/* Share Button */
+
+/* it was hidden directly inside the svg xml content */
+#share-btn-loading-icon {
+    display: none;
+}
+
+a {
+    text-decoration-line: underline;
+    font-weight: 600;
+}
+
+.animate-spin {
+    animation: spin 1s linear infinite;
+}
+
+@keyframes spin {
+    from {
+        transform: rotate(0deg);
+    }
+    to {
+        transform: rotate(360deg);
+    }
+}
+
+#share-btn-container {
+    display: flex;
+    padding-left: 0.5rem !important;
+    padding-right: 0.5rem !important;
+    background-color: #000000;
+    justify-content: center;
+    align-items: center;
+    border-radius: 9999px !important;
+    width: 15rem;
+}
+
+#share-btn {
+    all: initial;
+    color: #ffffff;
+    font-weight: 600;
+    cursor: pointer;
+    font-family: 'IBM Plex Sans', sans-serif;
+    margin-left: 0.5rem !important;
+    padding-top: 0.25rem !important;
+    padding-bottom: 0.25rem !important;
+}
+
+#share-btn * {
+    all: unset;
+}
+
+#share-btn-container div:nth-child(-n+2) {
+    width: auto !important;
+    min-height: 0px !important;
+}
+
+#share-btn-container .wrap {
+    display: none !important;
+}

utils.py

@@ -0,0 +1,45 @@
+import os
+from typing import List
+
+from settings import STATIC_PATH
+
+
+def get_file_as_string(file_name, path=STATIC_PATH) -> str:
+    """Loads the content of a file given its name
+    and returns all of its lines as a single string
+    if a file path is given, it will be used
+    instead of the default static path (from settings)
+
+    Args:
+        file_name (_type_): The name of the file to load.
+        path (str, optional): The path to the file. Defaults to the current directory.
+
+    Returns:
+        str: The content of the file as a single string
+    """
+    with open(os.path.join(path, file_name), mode="r", encoding="UTF-8") as f:
+        return f.read()
+
+
+def get_sections(string: str, delimiter: str, up_to: int = None) -> List[str]:
+    """Splits a string into sections given a delimiter
+
+    Args:
+        string (str): The string to split
+        delimiter (str): The delimiter to use
+        up_to (int, optional): The maximum number of sections to return.
+                Defaults to None (which means all sections)
+
+    Returns:
+        List[str]: The list of sections (up to the given limit, if any provided)
+    """
+    return [
+        section.strip()
+        for section in string.split(delimiter)
+        if (section and not section.isspace())
+    ][:up_to]
+
+
+def get_workers(safety: int = 4) -> int:
+    """Return the number of cores available on the current system, minus a safety margin."""
+    return max(1, os.cpu_count() - safety)