{"id":32750623,"url":"https://github.com/bogdanminko/laperf","last_synced_at":"2026-05-15T18:32:07.374Z","repository":{"id":319531549,"uuid":"1078485535","full_name":"bogdanminko/laperf","owner":"bogdanminko","description":"La Perf is a framework for AI performance benchmarking — covering LLMs, VLMs, embeddings, with power-metrics collection.","archived":false,"fork":false,"pushed_at":"2025-12-01T17:57:23.000Z","size":25527,"stargazers_count":2,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-05-01T20:36:56.879Z","etag":null,"topics":["ai-benchmark","ai-performance","apple-silicon","cuda","lmstudio","ml-benchmark","mlx","mps","nvidia-gpu","ollama","open-source-benchmark"],"latest_commit_sha":null,"homepage":"https://bogdanminko.github.io/laperf/index.html","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/bogdanminko.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":"docs/contributing.md","funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-10-17T20:11:08.000Z","updated_at":"2026-01-19T21:49:31.000Z","dependencies_parsed_at":"2025-10-19T11:37:23.005Z","dependency_job_id":null,"html_url":"https://github.com/bogdanminko/laperf","commit_stats":null,"previous_names":["bogdan01m/nobs","bogdanminko/nobs","bogdanminko/laperf"],"tags_count":3,"template":false,"template_full_name":null,"purl":"pkg:github/bogdanminko/laperf","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bogdanminko%2Flaperf","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bogdanminko%2Flaperf/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bogdanminko%2Flaperf/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bogdanminko%2Flaperf/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/bogdanminko","download_url":"https://codeload.github.com/bogdanminko/laperf/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bogdanminko%2Flaperf/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":33074840,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-15T11:35:32.926Z","status":"ssl_error","status_checked_at":"2026-05-15T11:35:31.362Z","response_time":103,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.5:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai-benchmark","ai-performance","apple-silicon","cuda","lmstudio","ml-benchmark","mlx","mps","nvidia-gpu","ollama","open-source-benchmark"],"created_at":"2025-11-03T23:01:08.162Z","updated_at":"2026-05-15T18:32:07.343Z","avatar_url":"https://github.com/bogdanminko.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\n\n# La Perf\n[![PyPI - laperf](https://img.shields.io/pypi/v/laperf?label=laperf\u0026color=blue)](https://pypi.org/project/laperf/)\n[![PyPI - laperf-power](https://img.shields.io/pypi/v/laperf-power?label=laperf-power\u0026color=blue)](https://pypi.org/project/laperf-power/)\n[![CUDA](https://img.shields.io/badge/CUDA-Supported-76B900?style=flat\u0026logo=nvidia\u0026logoColor=white)](https://developer.nvidia.com/cuda-zone)\n[![MPS](https://img.shields.io/badge/MPS-Optimized-000000?style=flat\u0026logo=apple\u0026logoColor=white)](https://developer.apple.com/metal/)\n[![MLX](https://img.shields.io/badge/MLX-Accelerated-FF6B35?style=flat\u0026logo=apple\u0026logoColor=white)](https://github.com/ml-explore/mlx)\n[![AI Performance](https://img.shields.io/badge/AI-Performance-FF6B6B?style=flat\u0026logo=tensorflow\u0026logoColor=white)](https://github.com/bogdanminko/laperf)\n[![Documentation](https://img.shields.io/badge/Documentation-GitHub%20Pages-2ECC71?style=flat\u0026logo=github\u0026logoColor=white)](https://bogdanminko.github.io/laperf/)\n\n### La Perf — a local AI performance benchmark\nfor comparing AI performance across different devices.\n\n\u003c/div\u003e\n\n---\nThe goal of this project is to create an all-in-one source of information you need **before buying your next laptop or PC for local AI tasks**.\n\nIt’s designed for **AI/ML engineers** who prefer to run workloads locally — and for **AI enthusiasts** who want to understand real-world device performance.\n\n\u003e **See full benchmark results here:**\n\u003e [Laperf Results](https://bogdanminko.github.io/laperf/results.html)\n\n![laperf-cli](assets/laperf_cli.png)\n\n## Table of Contents\n\n- [Overview](#overview)\n- [Philosophy](#philosophy)\n- [Benchmark Results](#benchmark-results)\n- [Quick Start](#-quick-start)\n- [Contributing](#-contributing)\n\n---\n\n## Overview\n### Tasks\nLa Perf is a collection of reproducible tests and community-submitted results for :\n\n- #### **Embeddings** — ✅ Ready (sentence-transformers, [IMDB dataset](https://huggingface.co/datasets/stanfordnlp/imdb))\n   sts models:\n   - [modernbert-embed-base](https://huggingface.co/nomic-ai/modernbert-embed-base)\n- #### **LLM inference** — ✅ Ready (LM Studio and Ollama, [Awesome Prompts dataset](https://huggingface.co/datasets/fka/awesome-chatgpt-prompts))\n   llm models:\n   - **LM Studio**: [gpt-oss-20b](https://lmstudio.ai/models/openai/gpt-oss-20b)\n     - *macOS*: `mlx-community/gpt-oss-20b-MXFP4-Q8` (MLX MXFP4-Q8)\n     - *Other platforms*: `lmstudio-community/gpt-oss-20b-GGUF` (GGUF)\n   - **Ollama**: [gpt-oss-20b](https://ollama.com/library/gpt-oss:20b)\n- #### **VLM inference** — ✅ Ready (LM Studio and Ollama, [Hallucination_COCO dataset](https://huggingface.co/datasets/DogNeverSleep/Hallucination_COCO))\n   vlm models:\n   - **LM Studio**: Qwen3-VL-8B-Thinking\n     - *macOS*: `mlx-community/Qwen3-VL-8B-Thinking-4bit` (MLX 4-bit)\n     - *Other platforms*: `Qwen/Qwen3-VL-8B-Thinking-GGUF-Q4_K_M` (Q4_K_M)\n   - **Ollama**: [qwen3-vl:8b](https://ollama.com/library/qwen3-vl:8b)\n      - **all platforms**: `qwen3-vl:8b` (Q4_K_M)\n- #### **Diffusion image generation** — 📋 Planned\n- #### **Speach to Text** - 📋 Planned (whisper)\n- #### **Classic ML** — 📋 Planned (scikit-learn, XGBoost, LightGBM, Catboost)\n\n**Note For mac-users**: If it's possible prefer to use lmstudio with `mlx` backend, which gives 10-20% more performance then `gguf`. If you run ollama (by default benchmarks runs both lmstudio and ollama) then you'll see a difference between `mlx` and `gguf` formats.\n\nThe `MLX` backend makes the benchmark harder to maintain, but it provides a more realistic performance view, since it’s easy to convert a `safetensors` model into an `mlx` x-bit model.\n\n### Requirements\n\nLa Perf is compatible with **Linux**, **macOS**, and **Windows**.\nFor embedding tasks, **8 GB of RAM** is usually sufficient.\nHowever for all tasks, it is **recommended to have at least 16 GB**, **18 GB** is better, and **24 GB or more** provides the best performance and reduces swap usage.\n\nIt’s designed to run anywhere the **`uv` package manager** is installed.\n\nIt’s recommended to use a GPU from **NVIDIA**, **AMD**, **Intel**, or **Apple**, since AI workloads run significantly faster on GPUs.\nMake sure to enable **full GPU offload** in tools like **LM Studio** or **Ollama** for optimal performance.\n\nFor embedding tasks, La Perf **automatically detects your available device** and runs computations accordingly.\n\n---\n\n## Benchmark Results\n\n\u003e **Last Updated**: 2025-11-19\n\n| Device | Platform | CPU | GPU | VRAM | Emb RPS P50 | LLM TPS P50 (lms) | LLM TPS P50 (ollama) | VLM TPS P50 (lms) | VLM TPS P50 (ollama) | GPU Power P50 | CPU Power P50 | Emb Efficiency (RPS/W) | LLM Efficiency (TPS/W) lms | LLM Efficiency (TPS/W) ollama | VLM Efficiency (TPS/W) lms | VLM Efficiency (TPS/W) ollama |\n|------|------|------|------|------|------|------|------|------|------|------|------|------|------|------|------|------|\n| ASUSTeK COMPUTER ASUS Vivobook Pro N6506MV | 🐧 Linux | Intel(R) Core(TM) Ultra 9 185H (16) | NVIDIA GeForce RTX 4060 Laptop GPU | 8 GB | 119.1 | 8.8 | 10.0 | 11.8 | 8.4 | 16.6 W | - | 7.18 | 0.53 | 0.60 | 0.71 | 0.51 |\n| Mac16,6 | 🍏 macOS | Apple M4 Max (14) | Apple M4 Max (32 cores) | shared with system RAM | 56.2 | 61.5 | 61.4 | 55.4 | 45.7 | 11.7 W | 1.0 W | 4.79 | 5.24 | 5.24 | 4.72 | 3.89 |\n| Mac16,6 (battery) | 🍏 macOS | Apple M4 Max (14) (battery) | Apple M4 Max (32 cores) (battery) | shared with system RAM | 56.2 | 59.1 | 60.6 | 54.8 | 44.9 | 11.4 W | 1.0 W | 4.94 | 5.21 | 5.33 | 4.83 | 3.95 |\n| OpenStack Nova 26.0.7-1 A100 40GB | 🐧 Linux | Intel(R) Xeon(R) Gold 6240R CPU @ 2.40GHz | NVIDIA A100-PCIE-40GB | 39 GB | 453.6 | - | 113.5 | - | 108.0 | 218.2 W | - | 2.08 | - | 0.52 | - | 0.50 |\n| OpenStack Nova A100 80GB | 🐧 Linux | Intel Xeon Processor (Icelake) | NVIDIA A100 80GB PCIe | 79 GB | 623.8 | - | 135.5 | - | 121.2 | 230.5 W | - | 2.71 | - | 0.59 | - | 0.53 |\n| OpenStack Nova RTX3090 | 🐧 Linux | Intel Xeon Processor (Cascadelake) | NVIDIA GeForce RTX 3090 | 24 GB | 349.5 | - | 114.8 | - | 105.3 | 345.6 W | - | 1.01 | - | 0.33 | - | 0.30 |\n| OpenStack Nova RTX4090 | 🐧 Linux | Intel Xeon Processor (Icelake) | NVIDIA GeForce RTX 4090 | 24 GB | 643.6 | - | 148.7 | - | 130.4 | 282.5 W | - | 2.28 | - | 0.53 | - | 0.46 |\n| OpenStack Nova Tesla T4 | 🐧 Linux | Intel Xeon Processor (Cascadelake) | Tesla T4 | 15 GB | 133.7 | - | 41.5 | - | 32.6 | 68.9 W | - | 1.94 | - | 0.60 | - | 0.47 |\n\n*RPS - Requests Per Second (embeddings throughput)*\n\n*TPS - Tokens Per Second (generation speed)*\n\n*W - Watts (power consumption)*\n\n*Efficiency metrics (RPS/W, TPS/W) are calculated using GPU power consumption*\n\n\n## ⚡ Quick Start\n\nFor a full quickstart and setup instructions, please visit the La Perf documentation: [Quickstart](https://bogdanminko.github.io/laperf/getting-started/quickstart.html).\n\n### 1. Clone the repository\n\n```bash\ngit clone https://github.com/bogdanminko/laperf.git\ncd laperf\n```\n\n### 2. (Optional) Configure environment variables\n\nLa Perf works out of the box with default settings, but you can customize it for different providers:\n\n```bash\ncp .env.example .env\n# Edit .env to change URLs, models, dataset sizes, etc.\n```\n\nSee [`.env.example`](.env.example) for all available options, including how to use custom OpenAI-compatible providers like vLLM, TGI, or LocalAI.\n\n### 3. Install dependencies (optional)\n\n```bash\nuv sync\n```\n\nThis will:\n\n- Create a virtual environment\n- Install all required dependencies\n- Set up the project for immediate use\n\n---\n\n## Running Your First Benchmark\n\n### Run all benchmarks\n**Using make**\n```bash\nmake bench\n```\n\n**Using uv**\n```bash\nuv run python main.py\n```\n\nThis will:\n\n1. **Auto-detect** your hardware (CUDA / MPS / CPU)\n2. **Run** all available benchmarks\n   (all are pre-selected — you can toggle individual ones in the TUI using `Space`)\n3. **Save** the results to `results/report_{your_device}.json`\n\n---\n\n## Power Monitoring Tool\n\nLa Perf includes a standalone real-time power monitoring tool that works independently from benchmarks.\n\n📦 **PyPI Package**: [laperf-power](https://pypi.org/project/laperf-power/)\n\n### Installation \u0026 Usage\n\n**Option 1: Run without installation (recommended)** ⭐\n```bash\n# Lightweight standalone package (~5 MB with psutil)\n# PyPI: https://pypi.org/project/laperf-power/\nuvx laperf-power\n\n# With custom options\nuvx laperf-power --interval 1.0 --output metrics.json\n```\n\n**Option 2: Install as a global tool**\n```bash\n# Lightweight standalone package\nuv tool install laperf-power\n# or: pip install laperf-power\n\n# Now available everywhere\nlaperf-power\nlaperf-power --interval 10.0 --no-sudo\n```\n\n**Option 3: Development mode (from source)**\n```bash\ngit clone https://github.com/bogdanminko/laperf.git\ncd laperf/laperf-power\nuv pip install -e .\nlaperf-power\n```\n\n### CLI Options\n\n```bash\nlaperf-power [OPTIONS]\n\nOptions:\n  --interval SECONDS    Sampling interval in seconds (default: 10.0)\n  --no-sudo            Disable sudo powermetrics on macOS\n  --output FILE        Save results to JSON file\n  -h, --help           Show help message\n```\n\n**Press Ctrl+C to stop and view statistics.**\n\n### What it monitors\n\n- **GPU**: Power (W), Utilization (%), VRAM (GB), Temperature (°C)\n- **CPU**: Power (W, macOS only with sudo), Utilization (%)\n- **System**: RAM usage (GB), Battery drain (%)\n\n### Example Output\n\n```\n⚡ REAL-TIME POWER MONITORING\n================================================================================\nStarted: 2025-11-27 14:30:00\nInterval: 1.0s\n================================================================================\n\nPress Ctrl+C to stop and view statistics\n\n[Sample #42] GPU: 11.7W 32% 8.2GB | CPU: 15% 1.0W | RAM: 16.3GB | Temp: 45°C\n```\n\n**Platform Support:**\n- **macOS**: Full support (with sudo for GPU/CPU power via `powermetrics`)\n- **Linux (NVIDIA)**: GPU metrics via `nvidia-smi`\n- **Windows**: Basic CPU/RAM metrics via `psutil`\n\n---\n\n## Running on GPU Servers (Docker)\n\nFor production deployments on cloud GPU instances or dedicated servers, you can use our Docker image:\n\n### Pull the image\n\n```bash\ndocker pull bogdan01m/laperf-cli:latest\n```\n\n### Run with NVIDIA GPU\n\n```bash\ndocker run --gpus all -it --rm \\\n  -v $(pwd)/results:/app/results \\\n  bogdan01m/laperf-cli:latest\n```\n\n### Run with AMD ROCm\n\n```bash\ndocker run --device=/dev/kfd --device=/dev/dri -it --rm \\\n  -v $(pwd)/results:/app/results \\\n  bogdan01m/laperf-cli:latest\n```\n\n### Run CPU-only\n\n```bash\ndocker run -it --rm \\\n  -v $(pwd)/results:/app/results \\\n  bogdan01m/laperf-cli:latest\n```\n\n**Note:** Results will be saved to the mounted `./results` directory on your host machine.\n\n---\n\n## Citation\n\nIf you use **LaPerf** in your research or reports, please cite it as follows:\n\n\u003e Minko B. (2025). *LaPerf: Local AI Performance Benchmark Suite.*\n\u003e GitHub repository. Available at: https://github.com/bogdan01m/laperf\n\u003e Licensed under the Apache License, Version 2.0.\n\n**BibTeX:**\n\n```bibtex\n@software{laperf,\n  author       = {Bogdan Minko},\n  title        = {LaPerf: Local AI Performance Benchmark Suite},\n  year         = {2025},\n  url          = {https://github.com/bogdan01m/laperf},\n  license      = {Apache-2.0},\n  note         = {GitHub repository}\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbogdanminko%2Flaperf","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fbogdanminko%2Flaperf","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbogdanminko%2Flaperf/lists"}