An official website of the United States government
In recent years, the pace of innovations in the fields of machine learning (ML) has accelerated, researchers in SysML have created algorithms and systems that parallelize ML training over multiple devices or computational nodes. As ML models become more structurally complex, many systems have struggled to provide all-round performance on a variety of models. Particularly, ML scale-up is usually underestimated in terms of the amount of knowledge and time required to map from an appropriate distribution strategy to the model. Applying parallel training systems to complex models adds nontrivial development overheads in addition to model prototyping, and often results in lower-than-expected performance. This tutorial identifies research and practical pain points in parallel ML training, and discusses latest development of algorithms and systems on addressing these challenges in both usability and performance. In particular, this tutorial presents a new perspective of unifying seemingly different distributed ML training strategies. Based on it, introduces new techniques and system architectures to simplify and automate ML parallelization. This tutorial is built upon the authors' years' of research and industry experience, comprehensive literature survey, and several latest tutorials and papers published by the authors and peer researchers. The tutorial consists of four parts. The first part will present a landscape of distributed ML training techniques and systems, and highlight the major difficulties faced by real users when writing distributed ML code with big model or big data. The second part dives deep to explain the mainstream training strategies, guided with real use case. By developing a new and unified formulation to represent the seemingly different data- and model- parallel strategies, we describe a set of techniques and algorithms to achieve ML auto-parallelization, and compiler system architectures for auto-generating and exercising parallelization strategies based on models and clusters. The third part of this tutorial exposes a hidden layer of practical pain points in distributed ML training: hyper-parameter tuning and resource allocation, and introduces techniques to improve these aspects. The fourth part is designed as a hands-on coding session, in which we will walk through the audiences on writing distributed training programs in Python, using the various distributed ML tools and interfaces provided by the Ray ecosystem.
more »
« less
RedCoast: A Lightweight Tool to Automate Distributed Training of LLMs on Any GPU/TPUs
The recent progress of AI can be largely attributed to large language models (LLMs). However, their escalating memory requirements introduce challenges for machine learning (ML) researchers and engineers. Addressing this requires developers to partition a large model to distribute it across multiple GPUs or TPUs. This necessitates considerable coding and intricate configuration efforts with existing model parallel tools, such as Megatron-LM, DeepSpeed, and Alpa. These tools require users’ expertise in machine learning systems (MLSys), creating a bottleneck in LLM development, particularly for developers without MLSys background. In this work, we present RedCoast (Redco), a lightweight and user-friendly tool crafted to automate distributed training and inference for LLMs, as well as to simplify ML pipeline development. The design of Redco emphasizes two key aspects. Firstly, to automate model parallelism, our study identifies two straightforward rules to generate tensor parallel strategies for any given LLM. Integrating these rules into Redco facilitates effortless distributed LLM training and inference, eliminating the need of additional coding or complex configurations. We demonstrate the effectiveness by applying Redco on a set of LLM architectures, such as GPT-J, LLaMA, T5, and OPT, up to the size of 66B. Secondly, we propose a mechanism that allows for the customization of diverse ML pipelines through the definition of merely three functions, avoiding redundant and formulaic code like multi-host related processing. This mechanism proves adaptable across a spectrum of ML algorithms, from foundational language modeling to complex algorithms like meta-learning and reinforcement learning. As a result, Redco implementations exhibit significantly fewer lines of code compared to their official counterparts. RedCoast (Redco) has been released under Apache 2.0 license at https://github.com/tanyuqian/redco.
more »
« less
- Award ID(s):
- 1955532
- PAR ID:
- 10540289
- Publisher / Repository:
- Association for Computational Linguistics
- Date Published:
- Journal Name:
- Proceedings of the conference Association for Computational Linguistics Meeting
- ISSN:
- 0736-587X
- Page Range / eLocation ID:
- 137 to 147
- Format(s):
- Medium: X
- Location:
- Mexico City, Mexico
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
The increasing use of machine learning and Large Language Models (LLMs) opens up opportunities to use these artificially intelligent algorithms in novel ways. This article proposes a methodology using LLMs to support traditional deductive coding in qualitative research. We began our analysis with three different sample texts taken from existing interviews. Next, we created a codebook and inputted the sample text and codebook into an LLM. We asked the LLM to determine if the codes were present in a sample text provided and requested evidence to support the coding. The sample texts were inputted 160 times to record changes between iterations of the LLM response. Each iteration was analogous to a new coder deductively analyzing the text with the codebook information. In our results, we present the outputs for these recursive analyses, along with a comparison of the LLM coding to evaluations made by human coders using traditional coding methods. We argue that LLM analysis can aid qualitative researchers by deductively coding transcripts, providing a systematic and reliable platform for code identification, and offering a means of avoiding analysis misalignment. Implications of using LLM in research praxis are discussed, along with current limitations.more » « less
-
Because of the naturalness of software and the rapid evolution of Machine Learning (ML) techniques, frequently repeated code change patterns (CPATs) occur often. They range from simple API migrations to changes involving several complex control structures such as for loops. While manually performing CPATs is tedious, the current state-of-the-art techniques for inferring transformation rules are not advanced enough to handle unseen variants of complex CPATs, resulting in a low recall rate. In this paper we present a novel, automated workflow that mines CPATs, infers the transformation rules, and then transplants them automatically to new target sites. We designed, implemented, evaluated and released this in a tool, PYEVOLVE. At its core is a novel data-flow, control-flow aware transformation rule inference engine. Our technique allows us to advance the state-of-the-art for transformation-by-example tools; without it, 70% of the code changes that PYEVOLVE transforms would not be possible to automate. Our thorough empirical evaluation of over 40,000 transformations shows 97% precision and 94% recall. By accepting 90% of CPATs generated by PYEVOLVE in famous open-source projects, developers confirmed its changes are useful.more » « less
-
Software developers often repeat the same code changes within a project or across different projects. These repetitive changes are known as “code change patterns” (CPATs). Automating CPATs is crucial to expedite the software development process. While current Transformation by Example (TBE) techniques can automate CPATs, they are limited by the quality and quantity of the provided input examples. Thus, they miss transforming code variations that do not have the exact syntax, data-, or control-flow of the provided input examples, despite being semantically similar. Large Language Models (LLMs), pre-trained on extensive source code datasets, offer a potential solution. Harnessing the capability of LLMs to generate semantically equivalent, yet previously unseen variants of the original CPAT could significantly increase the effectiveness of TBE systems. In this paper, we first discover best practices for harnessing LLMs to generate code variants that meet three criteria: correctness (semantic equivalence to the original CPAT), usefulness (reflecting what developers typically write), and applicability (aligning with the primary intent of the original CPAT). We then implement these practices in our tool PyCraft, which synergistically combines static code analysis, dynamic analysis, and LLM capabilities. By employing chain-of-thought reasoning, PyCraft generates variations of input examples and comprehensive test cases that identify correct variations with an F-measure of 96.6%. Our algorithm uses feedback iteration to expand the original input examples by an average factor of 58x. Using these richly generated examples, we inferred transformation rules and then automated these changes, resulting in an increase of up to 39x, with an average increase of 14x in target codes compared to a previous state-of-the-art tool that relies solely on static analysis. We submitted patches generated by PyCraft to a range of projects, notably esteemed ones like microsoft/DeepSpeed and IBM/inFairness. Their developers accepted and merged 83% the 86 CPAT instances submitted through 44 pull requests. This confirms the usefulness of these changes.more » « less
-
Few studies have compared Large Language Models (LLMs) to traditional Machine Learning (ML)-based automated scoring methods in terms of accuracy, ethics, and economics. Using a corpus of 1000 expert-scored and interview-validated scientific explanations derived from the ACORNS instrument, this study employed three LLMs and the ML-based scoring engine, EvoGrader. We measured scoring reliability (percentage agreement, kappa, precision, recall, F1), processing time, and explored contextual factors like ethics and cost. Results showed that with very basic prompt engineering, ChatGPT-4o achieved the highest performance across LLMs. Proprietary LLMs outperformed open-weight LLMs for most concepts. GPT-4o achieved robust but less accurate scoring than EvoGrader (~500 additional scoring errors). Ethical concerns over data ownership, reliability, and replicability over time were LLM limitations. EvoGrader offered superior accuracy, reliability, and replicability, but required, in its development a large, high-quality, human-scored corpus, domain expertise, and restricted assessment items. These findings highlight the diversity of considerations that should be used when considering LLM and ML scoring in science education. Despite impressive LLM advances, ML approaches may remain valuable in some contexts, particularly those prioritizing precision, reliability, replicability, privacy, and controlled implementation.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.18653/v1/2024.naacl-demo.14
