{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "
\n", "\n", "\n", "[![Documentation](https://img.shields.io/badge/Documentation-latest-blue.svg)](https://oumi.ai/docs/en/latest/index.html)\n", "[![Discord](https://img.shields.io/discord/1286348126797430814?label=Discord)](https://discord.gg/oumi)\n", "[![GitHub Repo stars](https://img.shields.io/github/stars/oumi-ai/oumi)](https://github.com/oumi-ai/oumi)\n", "\"Open\n", "
\n", "\n", "๐Ÿ‘‹ Welcome to Open Universal Machine Intelligence (Oumi)!\n", "\n", "๐Ÿš€ Oumi is a fully open-source platform that streamlines the entire lifecycle of foundation models - from [data preparation](https://oumi.ai/docs/en/latest/resources/datasets/datasets.html) and [training](https://oumi.ai/docs/en/latest/user_guides/train/train.html) to [evaluation](https://oumi.ai/docs/en/latest/user_guides/evaluate/evaluate.html) and [deployment](https://oumi.ai/docs/en/latest/user_guides/launch/launch.html). Whether you're developing on a laptop, launching large scale experiments on a cluster, or deploying models in production, Oumi provides the tools and workflows you need.\n", "\n", "๐Ÿค Make sure to join our [Discord community](https://discord.gg/oumi) to get help, share your experiences, and contribute to the project! If you are interested in joining one of the community's open-science efforts, check out our [open collaboration](https://oumi.ai/community) page.\n", "\n", "โญ If you like Oumi and you would like to support it, please give it a star on [GitHub](https://github.com/oumi-ai/oumi)." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Evaluation with Oumi\n", "\n", "This notebook provides a guide for running end-to-end evaluations on your trained model using Oumi. Specifically, it explores how the performance of LLaMA models evolves as we scale from 1B to 3B and 8B parameters." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Prerequisites and Environment\n", "\n", "โ—**NOTICE:** We recommend running this notebook on a GPU. If running on Google Colab, you can use the free T4 GPU runtime (Colab Menu: `Runtime` -> `Change runtime type`). Llama 8B is too large to fit on the T4 GPU for inference. You can remove it from the list of models to evaluate, or evaluate the SmolLM family of models instead of Llama in the `Experimental Setup` section.\n", "\n", "### Oumi Installation\n", "\n", "First, let's install Oumi, AlpacaEval, and vLLM. You can find more detailed instructions about Oumi installation [here](https://oumi.ai/docs/en/latest/get_started/installation.html)." ] }, { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Note: you may need to restart the kernel to use updated packages.\n" ] } ], "source": [ "%pip install oumi alpaca_eval vllm" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Llama Access\n", "\n", "Llama models are gated on HuggingFace Hub. To run this notebook, you must first complete the agreements for Llama [3.1](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct) and [3.2](https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct) on HuggingFace, and wait for them to be accepted. Then, specify `HF_TOKEN` below to enable access to the model if it's not already set.\n", "\n", "Usually, you can get the token by running this command `cat ~/.cache/huggingface/token` on your local machine." ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [], "source": [ "import os\n", "\n", "if not os.environ.get(\"HF_TOKEN\"):\n", " # NOTE: Set your Hugging Face token here if not already set.\n", " os.environ[\"HF_TOKEN\"] = \"\"\n", "\n", "hf_token = os.environ.get(\"HF_TOKEN\")\n", "print(f\"Using HF Token: '{hf_token}'\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### OpenAI Access\n", "\n", "AlpacaEval 2.0 calculates win rates by comparing a model's responses to reference responses. This process requires an annotator, with the default being [GPT-4 Turbo](https://github.com/tatsu-lab/alpaca_eval?tab=readme-ov-file#alpacaeval-20). To access the latest GPT-4 models, an OpenAI API key is necessary. You can find instructions for creating an OpenAI account and generating an API key on [OpenAI's quickstart webpage](https://platform.openai.com/docs/)." ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [], "source": [ "os.environ[\"OPENAI_API_KEY\"] = \"\" # Specify your OpenAI API key here" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "โš ๏ธ Cost considerations: The cost of running a standard AlpacaEval 2.0 evaluation and annotating 805 examples with GPT-4 Turbo is $3.50. You need to evaluate on the full dataset to reproduce the results presented in this notebook. However, if you are only interested in experimenting with the API, you can limit the evaluation to a smaller number of examples. For instance, annotating just 3 examples will cost less than 0.5ยข." ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [], "source": [ "# # The results shown below were generated from evaluating on the full dataset.\n", "# # Uncomment this if you want to reproduce the results.\n", "# NUM_ALPACA_EXAMPLES = None\n", "# NUM_MMLU_EXAMPLES = None\n", "\n", "NUM_ALPACA_EXAMPLES = 3 # Replace with None for full dataset evaluation.\n", "NUM_MMLU_EXAMPLES = 10 # Replace with None for full dataset evaluation." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Tutorial Directory Setup\n", "\n", "Finally, we will create a directory for the tutorial to store the results." ] }, { "cell_type": "code", "execution_count": 5, "metadata": {}, "outputs": [], "source": [ "from pathlib import Path\n", "\n", "tutorial_dir = \"./output/evaluation_tutorial/TEST_OUMI\"\n", "\n", "Path(tutorial_dir).mkdir(parents=True, exist_ok=True)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Experimental Setup\n", "\n", "In this experiment, we aim to compare the performance of models as we scale their size. For this purpose, we have selected LLaMA models with `1B`, `3B`, and `8B` parameters. In the code snippet below, you can specify any model hosted on HuggingFace, provide a path to a local directory containing your model, or use any other model format supported by Oumi inference. Additionally, note that we restrict the maximum number of tokens for each model to `8192` in order to optimize both cost and resource usage.\n" ] }, { "cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [], "source": [ "MODEL_NAMES = [\n", " \"meta-llama/Llama-3.2-1B-Instruct\",\n", " \"meta-llama/Llama-3.2-3B-Instruct\",\n", " \"meta-llama/Llama-3.1-8B-Instruct\",\n", "]\n", "MODEL_MAX_TOKENS = 8192\n", "\n", "# Alternative models to evaluate on Colab.\n", "# MODEL_NAMES = [\n", "# \"HuggingFaceTB/SmolLM-135M-Instruct\",\n", "# \"HuggingFaceTB/SmolLM-360M-Instruct\",\n", "# \"HuggingFaceTB/SmolLM-1.7B-Instruct\",\n", "# ]\n", "# MODEL_MAX_TOKENS = 2048" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "To conduct a comprehensive evaluation of a model's performance, it is essential to assess it across a [diverse set of benchmarks](https://oumi.ai/docs/en/latest/user_guides/evaluate/standardized_benchmarks.html#trade-offs). At a minimum, a [standardized benchmark](https://oumi.ai/docs/en/latest/user_guides/evaluate/standardized_benchmarks.html) is required to evaluate the model's knowledge and reasoning capabilities, while a [generative benchmark](https://oumi.ai/docs/en/latest/user_guides/evaluate/generative_benchmarks.html) is necessary to assess the quality of responses and its ability to follow instructions. In this evaluation, we select [MMLU Pro](https://arxiv.org/abs/2406.01574) for its focus on challenging, reasoning-intensive knowledge tasks and [AlpacaEval 2.0](https://arxiv.org/abs/2404.04475) for assessing the overall quality of the generated responses. To customize the evaluation tasks, please see [our documentation](https://oumi.ai/docs/en/latest/user_guides/evaluate/evaluation_config.html#configuration-options). " ] }, { "cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [], "source": [ "from oumi.core.configs import EvaluationTaskParams\n", "\n", "TASKS = [\n", " EvaluationTaskParams(\n", " evaluation_backend=\"lm_harness\",\n", " task_name=\"leaderboard_mmlu_pro\",\n", " num_samples=NUM_MMLU_EXAMPLES,\n", " ),\n", " EvaluationTaskParams(\n", " evaluation_backend=\"alpaca_eval\",\n", " num_samples=NUM_ALPACA_EXAMPLES,\n", " ),\n", "]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Evaluation\n", "\n", "To perform evaluations with Oumi, we define an [EvaluationConfig](https://oumi.ai/docs/en/latest/user_guides/evaluate/evaluation_config.html) for each model with the tasks discussed above, along with the relevant [model](https://github.com/oumi-ai/oumi/blob/main/src/oumi/core/configs/params/model_params.py) and [generation](https://github.com/oumi-ai/oumi/blob/main/src/oumi/core/configs/params/generation_params.py) parameters. If the system is equipped with a GPU, we strongly recommend configuring the [inference engine](https://oumi.ai/docs/en/latest/api/oumi.core.configs.html#oumi.core.configs.InferenceEngineType) to `VLLM` for optimal performance.\n", "\n", "The code snippet below illustrates how to run evaluations for our three models, storing the key metrics for each model in lists. Specifically, the length-controlled win rate for Alpaca is saved in `alpaca_lcwr`, while the accuracy for MMLU Pro is stored in `mmlu_acc`." ] }, { "cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [], "source": [ "from oumi.core.configs import (\n", " EvaluationConfig,\n", " GenerationParams,\n", " InferenceEngineType,\n", " ModelParams,\n", ")\n", "from oumi.evaluate import evaluate\n", "\n", "# Store the results for 1B, 3B, 8B models in the following lists.\n", "mmlu_acc = []\n", "alpaca_lcwr = []\n", "\n", "for model_name in MODEL_NAMES:\n", " print(f\"######################### Evaluating {model_name}...\")\n", " # Define the evaluation configuration.\n", " evaluation_config = EvaluationConfig(\n", " tasks=TASKS,\n", " model=ModelParams(\n", " model_name=model_name,\n", " model_max_length=MODEL_MAX_TOKENS,\n", " torch_dtype_str=\"bfloat16\",\n", " ),\n", " generation=GenerationParams(max_new_tokens=MODEL_MAX_TOKENS),\n", " inference_engine=InferenceEngineType.VLLM,\n", " output_dir=tutorial_dir,\n", " )\n", "\n", " # Evaluate the model.\n", " task_results = evaluate(evaluation_config)\n", "\n", " # Store the results.\n", " mmlu_acc.append(task_results[0][\"results\"][\"leaderboard_mmlu_pro\"][\"acc,none\"])\n", " alpaca_lcwr.append(\n", " task_results[1][\"results\"][\"alpaca_eval\"][\"length_controlled_winrate\"] / 100\n", " )" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Evaluation Results\n", "\n", "The evaluation results are shown below." ] }, { "cell_type": "code", "execution_count": 9, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "[0.20121343085106383, 0.281499335106383, 0.37142619680851063]\n", "[0.07650649505729805, 0.19105357406046689, 0.25572462470989504]\n" ] } ], "source": [ "print(mmlu_acc)\n", "print(alpaca_lcwr)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "When visualizing the results using `matplotlib`'s `pyplot`, we observe a clear trend: both the model's knowledge and reasoning abilities (measured by MMLU Pro) and the quality of its responses (measured by AlpacaEval 2.0) improve as the model size increases. A notable performance gain is observed when scaling from 1B to 3B parameters. However, beyond this, the rate of improvement begins to plateau, indicating diminishing returns with further increases in model size." ] }, { "cell_type": "code", "execution_count": 10, "metadata": {}, "outputs": [ { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "import matplotlib.pyplot as plt\n", "\n", "assert NUM_ALPACA_EXAMPLES is None and NUM_MMLU_EXAMPLES is None, (\n", " \"Skipping plotting as evaluating on such a small subset of data will not give \"\n", " \"meaningful results.\\nPlease set `NUM_ALPACA_EXAMPLES` and `NUM_MMLU_EXAMPLES` to \"\n", " \"None for full dataset evaluation.\"\n", ")\n", "\n", "# Data\n", "model_sizes = [1, 3, 8]\n", "plt.plot(model_sizes, mmlu_acc, label=\"MMLU Pro (Accuracy)\")\n", "plt.plot(model_sizes, alpaca_lcwr, label=\"AlpacaEval 2.0 (LC WR)\")\n", "\n", "# Show\n", "plt.xlabel(\"Model Size (Billion Parameters)\")\n", "plt.ylabel(\"Model Performance\")\n", "plt.legend()\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# ๐Ÿงญ What's Next?\n", "\n", "Congrats on finishing this notebook! Feel free to check out our other [notebooks](https://github.com/oumi-ai/oumi/tree/main/notebooks) in the [Oumi GitHub](https://github.com/oumi-ai/oumi), and give us a star! You can also join the Oumi community over on [Discord](https://discord.gg/oumi).\n", "\n", "๐Ÿ“ฐ Want to keep up with news from Oumi? Subscribe to our [Substack](https://blog.oumi.ai/) and [Youtube](https://www.youtube.com/@Oumi_AI)!\n", "\n", "โšก Interested in building custom AI in hours, not months? Apply to get [early access](https://oumi.ai/contact?utm_source=oumi_oss_tutorial_eval_oumi) to the Oumi Platform, or [chat with us](https://oumi.ai/book?utm_source=oumi_oss_tutorial_eval_oumi) to learn more!" ] } ], "metadata": { "kernelspec": { "display_name": "oumi", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.11.13" } }, "nbformat": 4, "nbformat_minor": 4 }