{ "cells": [ { "cell_type": "code", "execution_count": null, "id": "50532a6d-a3d7-485e-a2cc-76d70cadb0d2", "metadata": { "tags": [] }, "outputs": [], "source": [ "model_id = \"EleutherAI/gpt-j-6B\"\n", "revision = \"float16\" # use float16 weights to fit in 16GB GPUs" ] }, { "cell_type": "code", "execution_count": null, "id": "140cbbc0-9068-4605-ac29-4fe68f4e388a", "metadata": { "tags": [] }, "outputs": [], "source": [ "import ray" ] }, { "cell_type": "code", "execution_count": null, "id": "fa62af90-6652-4971-8ec1-8f1cac3fb01b", "metadata": { "tags": [] }, "outputs": [], "source": [ "ray.init(\n", " address=\"ray://example-cluster-kuberay-head-svc:10001\",\n", " runtime_env={\n", " \"pip\": [\n", " \"IPython\",\n", " \"boto3==1.26\",\n", " \"botocore==1.29\", \n", " \"datasets\",\n", " \"fastapi\",\n", " \"accelerate>=0.16.0\",\n", " \"transformers>=4.26.0\",\n", " \"numpy<1.24\", # remove when mlflow updates beyond 2.2\n", " \"torch\",\n", " ]\n", " }\n", ")" ] }, { "cell_type": "code", "execution_count": null, "id": "30bce290-51ec-4e8e-b6df-c179e8ebc281", "metadata": { "tags": [] }, "outputs": [], "source": [ "import pandas as pd\n", "\n", "from ray import serve\n", "from starlette.requests import Request\n", "\n", "\n", "@serve.deployment(ray_actor_options={\"num_gpus\": 1})\n", "class PredictDeployment:\n", " def __init__(self, model_id: str, revision: str = None):\n", " from transformers import AutoModelForCausalLM, AutoTokenizer\n", " import torch\n", "\n", " self.model = AutoModelForCausalLM.from_pretrained(\n", " model_id,\n", " revision=revision,\n", " torch_dtype=torch.float16,\n", " low_cpu_mem_usage=True,\n", " device_map=\"auto\", # automatically makes use of all GPUs available to the Actor\n", " )\n", " self.tokenizer = AutoTokenizer.from_pretrained(model_id)\n", "\n", " def generate(self, text: str) -> pd.DataFrame:\n", " input_ids = self.tokenizer(text, return_tensors=\"pt\").input_ids.to(\n", " self.model.device\n", " )\n", "\n", " gen_tokens = self.model.generate(\n", " input_ids,\n", " do_sample=True,\n", " temperature=0.9,\n", " max_length=100,\n", " )\n", " return pd.DataFrame(\n", " self.tokenizer.batch_decode(gen_tokens), columns=[\"responses\"]\n", " )\n", "\n", " async def __call__(self, http_request: Request) -> str:\n", " json_request: str = await http_request.json()\n", " prompts = []\n", " for prompt in json_request:\n", " text = prompt[\"text\"]\n", " if isinstance(text, list):\n", " prompts.extend(text)\n", " else:\n", " prompts.append(text)\n", " return self.generate(prompts)" ] }, { "cell_type": "code", "execution_count": null, "id": "33ea1b4a-afcf-4717-8ba7-3375331381ba", "metadata": { "tags": [] }, "outputs": [], "source": [ "deployment = PredictDeployment.bind(model_id=model_id, revision=revision)\n", "serve.run(deployment, host=\"0.0.0.0\")" ] }, { "cell_type": "code", "execution_count": null, "id": "bce2ad16-5248-4760-b92c-cb4517cb110c", "metadata": { "tags": [] }, "outputs": [], "source": [ "import requests\n", "\n", "prompt = (\n", " \"In a shocking finding, scientists discovered a herd of unicorns living in a remote, \"\n", " \"previously unexplored valley, in the Andes Mountains. Even more surprising to the \"\n", " \"researchers was the fact that the unicorns spoke perfect English.\"\n", ")\n", "\n", "sample_input = {\"text\": prompt}\n", "\n", "output = requests.post(\"http://example-cluster-kuberay-head-svc:8000/\", json=[sample_input]).json()\n", "print(output)" ] }, { "cell_type": "code", "execution_count": null, "id": "96f74370-2ac1-4dcd-81a5-e95cd2b8ac51", "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3 (ipykernel)", "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": "3.10.10" } }, "nbformat": 4, "nbformat_minor": 5 }