We recommend that you refer to https://github.com/baaivision/Emu3.5 if you need to perform model fine-tuning on NVIDIA hardware or deploy the model on physical robots.
Introduction
FlagOS is a unified heterogeneous computing software stack for large models, co-developed with leading global chip manufacturers. With core technologies such as the FlagScale distributed training/inference framework, FlagGems universal operator library, FlagCX communication library, and FlagTree unified compiler, the FlagRelease platform leverages the FlagOS stack to automatically produce and release various combinations of <chip + open-source model>. This enables efficient and automated model migration across diverse chips, opening a new chapter for large model deployment and application.
Based on this, the Emu3.5-FlagOS model is adapted for the Nvidia chip using the FlagOS software stack, enabling:
Out-of-the-box inference scripts with pre-configured hardware and software parameters
Released FlagOS container image supporting deployment within minutes
Consistency Validation
Rigorously evaluated through benchmark testing: Performance and results from the FlagOS software stack are compared against native stacks on multiple public.
Technical Overview
FlagScale Distributed Training and Inference Framework
FlagScale is an end-to-end framework for large models across heterogeneous computing resources, maximizing computational efficiency and ensuring model validity through core technologies. Its key advantages include:
Unified Deployment Interface: Standardized command-line tools support one-click service deployment across multiple hardware platforms, significantly reducing adaptation costs in heterogeneous environments.
Intelligent Parallel Optimization: Automatically generates optimal distributed parallel strategies based on chip computing characteristics, achieving dynamic load balancing of computation/communication resources.
Seamless Operator Switching: Deep integration with the FlagGems operator library allows high-performance operators to be invoked via environment variables without modifying model code.
FlagGems Universal Large-Model Operator Library
FlagGems is a Triton-based, cross-architecture operator library collaboratively developed with industry partners. Its core strengths include:
Full-stack Coverage: Over 100 operators, with a broader range of operator types than competing libraries.
High Efficiency: Employs unique code generation and runtime optimization techniques for faster secondary development and better runtime performance compared to alternatives.
FlagEval Evaluation Framework
FlagEval (Libra)** is a comprehensive evaluation system and open platform for large models launched in 2023. It aims to establish scientific, fair, and open benchmarks, methodologies, and tools to help researchers assess model and training algorithm performance. It features:
Multi-dimensional Evaluation: Supports 800+ model evaluations across NLP, CV, Audio, and Multimodal fields, covering 20+ downstream tasks including language understanding and image-text generation.
Industry-Grade Use Cases: Has completed horizontal evaluations of mainstream large models, providing authoritative benchmarks for chip-model performance validation.
Evaluation Results
Unlike other models, we judge, based on expert experience, that the capability levels demonstrated by CUDA and FlagOS's Emu3.5 are equivalent or not. In the four task domains of visual editing, text-to-image generation, visual guidance, and visual storytelling, a comprehensive judgment is made from multiple aspects including structure, image-text consistency, authenticity, spatial relationships, and style. The two sets of images are consistent in proportions and structure; their color schemes, material details, and texture distributions are similar, but the results from CUDA adhere more closely to real-world logic. Regarding overall painting or rendering styles, such as lighting methods, line characteristics, and detail density, the styles match. In localized precise editing, FlagOS exhibits more stable core features, whereas CUDA excels in maintaining texture and style continuity. Taking all dimensions into account, they can be considered to perform comparably overall.
You can find origin.tar and flagos.tar in the root directory of this model, which contain sample outputs from Emu3.5-CUDA and Emu3.5-FlagOS, respectively, across the four task domains.
1dockerexec -it flagos bash2cd /workspace/FlagScale
3python run.py --config-path examples/emu3.5/conf/ --config-name image_generation.yaml # to get image_generation results, like text to image4python run.py --config-path examples/emu3.5/conf/ --config-name interleaved_generation.yaml # to get interleaved_generation results, like visual_guidance
you can change /workspace/FlagScale/flagscale/inference/emu_utils/prompt_case.py to change the prompts.
