APOLLO
APOLLO: SGD-like Memory, AdamW-level Performance
Stars: 175
APOLLO is a memory-efficient optimizer designed for large language model (LLM) pre-training and full-parameter fine-tuning. It offers SGD-like memory cost with AdamW-level performance. The optimizer integrates low-rank approximation and optimizer state redundancy reduction to achieve significant memory savings while maintaining or surpassing the performance of Adam(W). Key contributions include structured learning rate updates for LLM training, approximated channel-wise gradient scaling in a low-rank auxiliary space, and minimal-rank tensor-wise gradient scaling. APOLLO aims to optimize memory efficiency during training large language models.
README:
APOLLO is archievd at Figshare with DOI: https://doi.org/10.6084/m9.figshare.28558319.v1 for MLSys artifact evaluation.
A memory-efficient optimizer designed for large language model (LLM) pre-training and full-parameter fine-tuning, offering SGD-like memory cost with AdamW-level performance.
🔗 Paper • Project Page • Video • Hugging Face Transformers • LLaMA-Factory • FluxML • Hacker News
-
[2025/2] APOLLO is accepted to MLSys 2025, cheers!
-
[2025/2] APOLLO is integrated into Hugging Face Transformers. Try APOLLO with Hugging Face trainer!
-
[2025/2] Fix the compatibility issue of quantized APOLLO with bitsandbytes.
-
[2025/1] APOLLO is integrated into LLaMA-Factory. Try it for memory-efficient LLM full-parameter fine-tuning.
-
[2024/12] We are happy to release the official implementation of APOLLO v1.0.0 in PyPI (see here). We support QAPOLLO using int8 weight quantization from Q-Galore.
-
[2024/12] APOLLO validated by third-party Julia implementation!: Our APOLLO optimizer has been independently validated by a third party using a Julia implementation. Check out the post. They are also working to integrate APOLLO into FluxML.
-
[2024/12] APOLLO Paper Released: Our paper is now available on arXiv! Check it out here: [Paper].
We introduce APOLLO (Approximated Gradient Scaling for Memory Efficient LLM Optimization), a novel method designed to optimize the memory efficiency of training large language models (LLM), offering SGD-like memory cost while delivering AdamW-level performance for both pre-training and finetuning!
APOLLO effectively integrates two major ideas for memory-efficient LLM training: low-rank approximation (GaLore) and optimizer state redundancy reduction (Adam-mini). However, APOLLO takes memory efficiency to a new level, achieving significant memory savings (below GaLore and its variants, and close to SGD) while maintaining or surpassing the performance of Adam(W).
Our key contributions include:
-
Structured Learning Rate Updates for LLM Training: We identify that structured learning rate updates, such as channel-wise or tensor-wise scaling, are sufficient for LLM training. This approach explores redundancy in AdamW's element-wise learning rate update rule, forming a basis for our APOLLO method.
-
Approximated Channel-wise Gradient Scaling in a Low-Rank Auxiliary Space (APOLLO):
APOLLO proposes a practical and memory-efficient method to approximate channel-wise gradient scaling factors in an auxiliary low-rank space using pure random projections. This method achieves superior performance compared to AdamW, even with lower-rank approximations, while maintaining excellent memory efficiency. -
Minimal-Rank Tensor-wise Gradient Scaling (APOLLO-Mini):
APOLLO-Mini introduces extreme memory efficiency by applying tensor-wise gradient scaling using only a rank-1 auxiliary sub-space. This results in SGD-level memory costs while outperforming AdamW, showcasing the effectiveness of the approach.
Figure 1: The APOLLO Framework for Memory-Efficient LLM Training. The channel-wise or tensor-wise gradient scaling factor is obtained via an auxiliary low-rank optimizer state, constructed using pure random projection (no SVD required).
Figure 2: System Benefits of APOLLO for Pre-training LLaMA 7B. (left): Memory breakdown comparison for a single batch size; (right): End-to-end training throughput on 8 A100-80GB GPUs
You can install the APOLLO optimizer directly from pip:
pip install apollo-torchTo install APOLLO from the source code:
git clone https://github.com/zhuhanqing/APOLLO.git
cd APOLLO
pip install -e .pip install -r exp_requirements.txtfrom apollo_torch import APOLLOAdamW
# define param groups as lowrank_params and regular params
param_groups = [{'params': non_lowrank_params},
{'params':
lowrank_params,
'rank': 1,
'proj': 'random',
'scale_type': 'tensor',
'scale': 128,
'update_proj_gap': 200,
'proj_type': 'std'}]
optimizer = APOLLO(param_groups, lr=0.01)
For APOLLO and APOLLO-Mini, we have the following arguments
- Specifies the rank of the auxiliary sub-space used for gradient scaling.
-
Default value:
-
256for APOLLO works well for 1B and 7B model. -
1for APOLLO-Mini.
-
- Determines how the scaling factors are applied:
-
channel: Applies gradient scaling at the channel level (APOLLO) -
tensor: Applies gradient scaling at the tensor level (APOLLO-Mini).
-
The scale parameter plays a crucial role in heuristically adjusting gradient updates to compensate for scaling factor approximation errors arising from the use of a lower rank. Proper tuning of this parameter can significantly improve performance:
-
1: Default value for APOLLO (validated on A100 GPUs). -
128: Default value for APOLLO-Mini. For larger models, experimenting with higher values is recommended.
To stabilize training, we adopt the Norm-Growth Limiter (NL) from Fira, which has shown to be slightly more effective than traditional gradient clipping.
There are two ways to apply the Norm-Growth Limiter based on when it's used relative to the heuristical (scale):
-
After Scaling: NL is applied after the gradient is multiplied by the
scale.- Recommended for when training involves fewer warmup steps, e.g., LLaMA 60M and 130M with APOLLO-Mini.
- Enable this by setting
--scale_front.
-
Before Scaling: NL is applied before the gradient is scaled.
- With sufficient warmup steps, both methods yield similar performance for large models.
We provide the command in scripts/benchmark_c4 for pretraining LLaMA model with sizes from 60M to 7B on C4 dataset.
# num_rank: 1 for APOLLO-Mini, 1/4 of the original dim for APOLLO (same as Galore)
# scale_type: channel or tensor
# projection type: random (option: svd)
# scale: related with rank, larger rank generally works well with smaller scale, we use 128 for rank=1
Compared to academic settings, the industry trains large language models (LLMs) with significantly longer context windows (1k-8k tokens) and on hundreds of billions of tokens.
Accordingly, we further validate the effectiveness of the APOLLO series by pre-training a LLaMA-350M on a 1024-token context window—four times larger than the original GaLore usage. To establish a robust baseline, we vary AdamW’s learning rate across [1e-3, 2.5e-3, 5e-3, 7.5e-3, 1e-2]. We also “lazily” tune the scale factor of the APOLLO series by testing APOLLO in [√1, √2, √3] and APOLLO-Mini in [√128, √256, √384], while keeping the learning rate fixed at 1e-2.
Both APOLLO and APOLLO-Mini demonstrate superior performance compared to AdamW, while drastically reducing optimizer memory usage—by as much as 1/8 or even 1/1024 of AdamW’s requirements. Moreover, these methods tend to exhibit even stronger performance in later stages, when more training tokens are involved. This makes them a highly promising option for partial LLM pre-training scenarios involving long context windows and trillions of training tokens.
Figure 3: Perplexity curves of the LLaMA-350M model trained in a long-context window setting.
The command of training LLaMA-7B model on single GPU as provided within scripts/single_gpu. With 1 batch size, the following scripts can pre-train a LLaMA-7B model within 11GB memory (tested on a single A100 GPU)
Now we support APOLLO in LLaMA-Factory. We have added a test in the examples/extras/apollo directory.
We conducted a comparative evaluation with GaLore by fine-tuning models and testing on the MMLU task.
Average: 64.96
STEM: 55.43
Social Sciences: 75.66
Humanities: 59.72
Other: 71.25
With a scaling factor derived from the ratio of LLaMA-8B dimension (4096) to rank (128):
Average: 65.03
STEM: 55.47
Social Sciences: 76.15
Humanities: 59.60
Other: 71.28
Large language models (LLMs) demonstrate remarkable capabilities but are notoriously memory-intensive during training, particularly with the popular Adam optimizer. This memory burden often necessitates using more GPUs, smaller batch sizes, or high-end hardware, thereby limiting scalability and training efficiency. To address this, various memory-efficient optimizers have been proposed to reduce optimizer memory usage. However, they face key challenges: (i) reliance on costly SVD operations (e.g., GaLore, Fira); (ii) significant performance trade-offs compared to AdamW (e.g., Flora); and (iii) still substantial memory overhead of optimization states in order to maintain competitive performance (e.g., 1/4 rank in Galore, and full-rank first momentum in Adam-mini).
In this work, we investigate the redundancy in Adam's learning rate adaption rule and identify that it can be coarsened as a structured learning rate update (channel-wise or tensor-wise). Based on this insight, we propose a novel approach, Approximated Gradient Scaling for Memory Efficient LLM Optimization (APOLLO), which approximate the channel-wise learning rate scaling with an auxiliary low-rank optimizer state based on pure random projection. The structured learning rate update rule makes APOLLO highly tolerant to further memory reduction with lower rank, halving the rank while delivering similar pre-training performance. We further propose an extreme memory-efficient version, APOLLO-mini, which utilizes tensor-wise scaling with only a rank-1 auxiliary sub-space, achieving SGD-level memory cost but superior pre-training performance than Adam(W).
We conduct extensive experiments across different tasks and model architectures, showing that APOLLO series performs generally on-par with, or even better than Adam(W). Meanwhile, APOLLO achieves even greater memory savings than Galore, by almost eliminating the optimization states in AdamW. These savings translate into significant system benefits:
- Enhanced Throughput: APOLLO and APOLLO-mini achieve up to 3x throughput on a 4xA100-80GB setup compared to Adam by fully utilizing memory to support 4x larger batch sizes.
- Improved Model Scalability: APOLLO-mini for the first time enables pre-training LLaMA-13B model with naive DDP on A100-80G without requiring other system-level optimizations
- Low-End GPU Pre-training: Combined with quantization, the APOLLO series for the first time enables the training of LLaMA-7B from scratch using less than 12 GB of memory.
- [ ] Support APOLLO with FSDP.
For questions or collaboration inquiries, feel free to reach out our core contributors:
- 📧 Email: [email protected]
- 📧 Email: [email protected]
If you find APOLLO useful in your work, please consider citing our paper:
@misc{zhu2024apollosgdlikememoryadamwlevel,
title={APOLLO: SGD-like Memory, AdamW-level Performance},
author={Hanqing Zhu and Zhenyu Zhang and Wenyan Cong and Xi Liu and Sem Park and Vikas Chandra and Bo Long and David Z. Pan and Zhangyang Wang and Jinwon Lee},
year={2024},
eprint={2412.05270},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2412.05270},
}The majority of APOLLO is licensed under CC-BY-NC, however portions of the project are available under separate license terms: GaLore is licensed under the Apache 2.0 license.
For Tasks:
Click tags to check more tools for each tasksFor Jobs:
Alternative AI tools for APOLLO
Similar Open Source Tools
APOLLO
APOLLO is a memory-efficient optimizer designed for large language model (LLM) pre-training and full-parameter fine-tuning. It offers SGD-like memory cost with AdamW-level performance. The optimizer integrates low-rank approximation and optimizer state redundancy reduction to achieve significant memory savings while maintaining or surpassing the performance of Adam(W). Key contributions include structured learning rate updates for LLM training, approximated channel-wise gradient scaling in a low-rank auxiliary space, and minimal-rank tensor-wise gradient scaling. APOLLO aims to optimize memory efficiency during training large language models.
eole
EOLE is an open language modeling toolkit based on PyTorch. It aims to provide a research-friendly approach with a comprehensive yet compact and modular codebase for experimenting with various types of language models. The toolkit includes features such as versatile training and inference, dynamic data transforms, comprehensive large language model support, advanced quantization, efficient finetuning, flexible inference, and tensor parallelism. EOLE is a work in progress with ongoing enhancements in configuration management, command line entry points, reproducible recipes, core API simplification, and plans for further simplification, refactoring, inference server development, additional recipes, documentation enhancement, test coverage improvement, logging enhancements, and broader model support.
MM-RLHF
MM-RLHF is a comprehensive project for aligning Multimodal Large Language Models (MLLMs) with human preferences. It includes a high-quality MLLM alignment dataset, a Critique-Based MLLM reward model, a novel alignment algorithm MM-DPO, and benchmarks for reward models and multimodal safety. The dataset covers image understanding, video understanding, and safety-related tasks with model-generated responses and human-annotated scores. The reward model generates critiques of candidate texts before assigning scores for enhanced interpretability. MM-DPO is an alignment algorithm that achieves performance gains with simple adjustments to the DPO framework. The project enables consistent performance improvements across 10 dimensions and 27 benchmarks for open-source MLLMs.
MathVerse
MathVerse is an all-around visual math benchmark designed to evaluate the capabilities of Multi-modal Large Language Models (MLLMs) in visual math problem-solving. It collects high-quality math problems with diagrams to assess how well MLLMs can understand visual diagrams for mathematical reasoning. The benchmark includes 2,612 problems transformed into six versions each, contributing to 15K test samples. It also introduces a Chain-of-Thought (CoT) Evaluation strategy for fine-grained assessment of output answers.
RAGEN
RAGEN is a reinforcement learning framework designed to train reasoning-capable large language model (LLM) agents in interactive, stochastic environments. It addresses challenges such as multi-turn interactions and stochastic environments through a Markov Decision Process (MDP) formulation, Reason-Interaction Chain Optimization (RICO) algorithm, and progressive reward normalization strategies. The framework enables LLMs to reason and interact with the environment, optimizing entire trajectories for long-horizon reasoning while maintaining computational efficiency.
Simplifine
Simplifine is an open-source library designed for easy LLM finetuning, enabling users to perform tasks such as supervised fine tuning, question-answer finetuning, contrastive loss for embedding tasks, multi-label classification finetuning, and more. It provides features like WandB logging, in-built evaluation tools, automated finetuning parameters, and state-of-the-art optimization techniques. The library offers bug fixes, new features, and documentation updates in its latest version. Users can install Simplifine via pip or directly from GitHub. The project welcomes contributors and provides comprehensive documentation and support for users.
DevoxxGenieIDEAPlugin
Devoxx Genie is a Java-based IntelliJ IDEA plugin that integrates with local and cloud-based LLM providers to aid in reviewing, testing, and explaining project code. It supports features like code highlighting, chat conversations, and adding files/code snippets to context. Users can modify REST endpoints and LLM parameters in settings, including support for cloud-based LLMs. The plugin requires IntelliJ version 2023.3.4 and JDK 17. Building and publishing the plugin is done using Gradle tasks. Users can select an LLM provider, choose code, and use commands like review, explain, or generate unit tests for code analysis.
verl
veRL is a flexible and efficient reinforcement learning training framework designed for large language models (LLMs). It allows easy extension of diverse RL algorithms, seamless integration with existing LLM infrastructures, and flexible device mapping. The framework achieves state-of-the-art throughput and efficient actor model resharding with 3D-HybridEngine. It supports popular HuggingFace models and is suitable for users working with PyTorch FSDP, Megatron-LM, and vLLM backends.
MARBLE
MARBLE (Multi-Agent Coordination Backbone with LLM Engine) is a modular framework for developing, testing, and evaluating multi-agent systems leveraging Large Language Models. It provides a structured environment for agents to interact in simulated environments, utilizing cognitive abilities and communication mechanisms for collaborative or competitive tasks. The framework features modular design, multi-agent support, LLM integration, shared memory, flexible environments, metrics and evaluation, industrial coding standards, and Docker support.
TaskingAI
TaskingAI brings Firebase's simplicity to **AI-native app development**. The platform enables the creation of GPTs-like multi-tenant applications using a wide range of LLMs from various providers. It features distinct, modular functions such as Inference, Retrieval, Assistant, and Tool, seamlessly integrated to enhance the development process. TaskingAI’s cohesive design ensures an efficient, intelligent, and user-friendly experience in AI application development.
petals
Petals is a tool that allows users to run large language models at home in a BitTorrent-style manner. It enables fine-tuning and inference up to 10x faster than offloading. Users can generate text with distributed models like Llama 2, Falcon, and BLOOM, and fine-tune them for specific tasks directly from their desktop computer or Google Colab. Petals is a community-run system that relies on people sharing their GPUs to increase its capacity and offer a distributed network for hosting model layers.
chat-with-mlx
Chat with MLX is an all-in-one Chat Playground using Apple MLX on Apple Silicon Macs. It provides privacy-enhanced AI for secure conversations with various models, easy integration of HuggingFace and MLX Compatible Open-Source Models, and comes with default models like Llama-3, Phi-3, Yi, Qwen, Mistral, Codestral, Mixtral, StableLM. The tool is designed for developers and researchers working with machine learning models on Apple Silicon.
veScale
veScale is a PyTorch Native LLM Training Framework. It provides a set of tools and components to facilitate the training of large language models (LLMs) using PyTorch. veScale includes features such as 4D parallelism, fast checkpointing, and a CUDA event monitor. It is designed to be scalable and efficient, and it can be used to train LLMs on a variety of hardware platforms.
llmc
llmc is an off-the-shell tool designed for compressing LLM, leveraging state-of-the-art compression algorithms to enhance efficiency and reduce model size without compromising performance. It provides users with the ability to quantize LLMs, choose from various compression algorithms, export transformed models for further optimization, and directly infer compressed models with a shallow memory footprint. The tool supports a range of model types and quantization algorithms, with ongoing development to include pruning techniques. Users can design their configurations for quantization and evaluation, with documentation and examples planned for future updates. llmc is a valuable resource for researchers working on post-training quantization of large language models.
kollektiv
Kollektiv is a Retrieval-Augmented Generation (RAG) system designed to enable users to chat with their favorite documentation easily. It aims to provide LLMs with access to the most up-to-date knowledge, reducing inaccuracies and improving productivity. The system utilizes intelligent web crawling, advanced document processing, vector search, multi-query expansion, smart re-ranking, AI-powered responses, and dynamic system prompts. The technical stack includes Python/FastAPI for backend, Supabase, ChromaDB, and Redis for storage, OpenAI and Anthropic Claude 3.5 Sonnet for AI/ML, and Chainlit for UI. Kollektiv is licensed under a modified version of the Apache License 2.0, allowing free use for non-commercial purposes.
voice-pro
Voice-Pro is an integrated solution for subtitles, translation, and TTS. It offers features like multilingual subtitles, live translation, vocal remover, and supports OpenAI Whisper and Open-Source Translator. The tool provides a Studio tab for various functions, Whisper Caption tab for subtitle creation, Translate tab for translation, TTS tab for text-to-speech, Live Translation tab for real-time voice recognition, and Batch tab for processing multiple files. Users can download YouTube videos, improve voice recognition accuracy, create automatic subtitles, and produce multilingual videos with ease. The tool is easy to install with one-click and offers a Web-UI for user convenience.
For similar tasks
APOLLO
APOLLO is a memory-efficient optimizer designed for large language model (LLM) pre-training and full-parameter fine-tuning. It offers SGD-like memory cost with AdamW-level performance. The optimizer integrates low-rank approximation and optimizer state redundancy reduction to achieve significant memory savings while maintaining or surpassing the performance of Adam(W). Key contributions include structured learning rate updates for LLM training, approximated channel-wise gradient scaling in a low-rank auxiliary space, and minimal-rank tensor-wise gradient scaling. APOLLO aims to optimize memory efficiency during training large language models.
For similar jobs
sweep
Sweep is an AI junior developer that turns bugs and feature requests into code changes. It automatically handles developer experience improvements like adding type hints and improving test coverage.
teams-ai
The Teams AI Library is a software development kit (SDK) that helps developers create bots that can interact with Teams and Microsoft 365 applications. It is built on top of the Bot Framework SDK and simplifies the process of developing bots that interact with Teams' artificial intelligence capabilities. The SDK is available for JavaScript/TypeScript, .NET, and Python.
ai-guide
This guide is dedicated to Large Language Models (LLMs) that you can run on your home computer. It assumes your PC is a lower-end, non-gaming setup.
classifai
Supercharge WordPress Content Workflows and Engagement with Artificial Intelligence. Tap into leading cloud-based services like OpenAI, Microsoft Azure AI, Google Gemini and IBM Watson to augment your WordPress-powered websites. Publish content faster while improving SEO performance and increasing audience engagement. ClassifAI integrates Artificial Intelligence and Machine Learning technologies to lighten your workload and eliminate tedious tasks, giving you more time to create original content that matters.
chatbot-ui
Chatbot UI is an open-source AI chat app that allows users to create and deploy their own AI chatbots. It is easy to use and can be customized to fit any need. Chatbot UI is perfect for businesses, developers, and anyone who wants to create a chatbot.
BricksLLM
BricksLLM is a cloud native AI gateway written in Go. Currently, it provides native support for OpenAI, Anthropic, Azure OpenAI and vLLM. BricksLLM aims to provide enterprise level infrastructure that can power any LLM production use cases. Here are some use cases for BricksLLM: * Set LLM usage limits for users on different pricing tiers * Track LLM usage on a per user and per organization basis * Block or redact requests containing PIIs * Improve LLM reliability with failovers, retries and caching * Distribute API keys with rate limits and cost limits for internal development/production use cases * Distribute API keys with rate limits and cost limits for students
uAgents
uAgents is a Python library developed by Fetch.ai that allows for the creation of autonomous AI agents. These agents can perform various tasks on a schedule or take action on various events. uAgents are easy to create and manage, and they are connected to a fast-growing network of other uAgents. They are also secure, with cryptographically secured messages and wallets.
griptape
Griptape is a modular Python framework for building AI-powered applications that securely connect to your enterprise data and APIs. It offers developers the ability to maintain control and flexibility at every step. Griptape's core components include Structures (Agents, Pipelines, and Workflows), Tasks, Tools, Memory (Conversation Memory, Task Memory, and Meta Memory), Drivers (Prompt and Embedding Drivers, Vector Store Drivers, Image Generation Drivers, Image Query Drivers, SQL Drivers, Web Scraper Drivers, and Conversation Memory Drivers), Engines (Query Engines, Extraction Engines, Summary Engines, Image Generation Engines, and Image Query Engines), and additional components (Rulesets, Loaders, Artifacts, Chunkers, and Tokenizers). Griptape enables developers to create AI-powered applications with ease and efficiency.