search

cli99 / llm-analysis

Latency and Memory Analysis of Transformer Models for Training and Inference
Apache License 2.0
356 stars 42 forks source link
analysis deep-learning language-model language-models machine-learning nlp transformers

readme

llm-analysis

PyPI Read the Docs Tests Codecov GitHub license

Latency and Memory Analysis of Transformer Models for Training and Inference

Overview

Many formulas or equations are floating around in papers, blogs, etc., about how to calculate training or inference latency and memory for Large Language Models (LLMs) or Transformers. Rather than doing math on papers or typing in Excel sheets, let's automate the boring stuff with llm-analysis :gear:!

Given the specified model, GPU, data type, and parallelism configurations, llm-analysis estimates the latency and memory usage of LLMs for training or inference. With llm-analysis, one can easily try out different training/inference setups theoretically, and better understand the system performance for different scenarios.

llm-analysis helps answer questions such as:

Examples

Check the example use cases. With llm-analysis, you can do such analysis in minutes :rocket:!

Quick Start

Using the LLMAnalysis class

To integrate llm-analysis in your code, use the LLMAnalysis class. Refer to doc LLMAnalysis for details.

LLMAnalysis is constructed with flops and memory efficiency numbers and the following configuration classes:

Then LLMAnalysis can be queried with different arguments through the training and inference methods.

Using the Entry Point Functions for Command Line

llm-analysis provides two entry functions, train and infer, for ease of use through the command line interface. Run

python -m llm_analysis.analysis train --help

or

python -m llm_analysis.analysis infer --help

to check the options or read the linked doc. Refer to the examples to see how they are used.

train and infer use the pre-defined name-to-configuration mappings (model_configs, gpu_configs, dtype_configs) and other user-input arguments to construct the LLMAnalysis and do the query.

The pre-defined mappings are populated at the runtime from the model, GPU, and data type configuration json files under model_configs, gpu_configs, and dtype_configs. To add a new model, GPU or data type to the mapping for query, just add a json description file to the corresponding folder.

llm-analysis also supports retrieving ModelConfig from a model config json file path or Hugging Face with the model name .

A list of handy commands is provided to query against the pre-defined mappings as well as Hugging Face, or to dump configurations. Run python -m llm_analysis.config --help for details.

Some examples:

python -m llm_analysis.config get_model_config_by_name EleutherAI/gpt-neox-20b

gets the ModelConfig from the populated mapping by name, if not found, llm-analysis tries to get it from HuggingFace.

Note that LLaMA models need at least transformers-4.28.1 to retrieve, either update to a later transformers library, or use the predefined ModelConfig for LLaMA models (/ in model names are replaced with _).

python -m llm_analysis.config list_gpu_configs

lists the names of all predefined GPU configurations, then you can query with

python -m llm_analysis.config get_gpu_config_by_name a100-sxm-80gb

to show the corresponding GPUConfig.

How to Set FLOPS and Memory Efficiency

Setting flops and memory efficiency to 1 (default) gives the lower bound of training or inference latency, as it assumes the peak hardware performance (which is never the case). A close-to-reality flops or memory efficiency can be found by benchmarking and profiling using the input dimensions in the model.

If one has to make assumptions, for flops efficiency, literature reports up to 0.5 for large scale model training, and up to 0.7 for inference; 0.9 can be an aggressive target for memory efficiency.

Current Scope and Limitations

llm-analysis aims to provide a lower-bound estimation of memory usage and latency.

Parallelism Scheme

llm-analysis currently covers Tensor Parallelism (tp), Pipeline Parallelism (pp), Sequence Parallelism (sp), Expert Parallelism (ep), and Data Parallelism (dp).

Communication

tp communication is calculated as using ring allreduce. ep communication is calculated as using alltoall. dp communication time to unshard model weight when using FSDP or DeepSpeed ZeRO is estimated and compared against the compute latency, the larger value of the two is used for the overall latency. Other dp and pp communications are ignored for now, i.e. assuming perfect computation and communication overlapping, which is not true when communication cannot overlap with compute due to dependency, or when communication is too long to hide due to slow interconnect or large data volume.

Activation Recomputation

llm-analysis supports both full and selective activation recomputation.

activation_recomputation what is checkpointed and recomputed
0 No activation recomputation; requires the most amount of memory
1 Checkpoints the attention computation (QK^T matrix multiply, softmax, softmax dropout, and attention over V.) in the attention module of a transformer layer; as described in Reducing Activation Recomputation in Large Transformer Models.
2 Checkpoints the input to the attention module in a transformer layer; requires an extra forward pass on attention.
3 Checkpoints the input to the sequence of modules (layernom-attention-layernom) in a transformer layer; requires an extra forward pass on (layernom-attention-layernom).
4 Full activation recomputation stores the input to the transformer layer; requires the least amount of memory; requires an extra forward pass of the entire layer.

Data Types

Data types are expressed with the number of bits, only 32 (FP32, TF32), 16 (FP16, BF16), 8 (INT8), and 4 (INT4) bits data types are modeled for now.

Fine-Tuning

Fine-tuning is modeled the same (controlled by total_num_tokens passed to the train entry function) as pre-training, thus assuming full (all model parameters) fine-tuning. Parameter-efficient fine-tuning (PEFT) is in future support.

Assumptions in Inference

Inference assumes perfect overlapping of compute and memory operations when calculating latency, and maximum memory reuse when calculating memory usage.

TODOs (stay tuned :radio:)

Check the TODOs below for what's next and stay tuned :radio:! Any contributions or feedback are highly welcome!

Citation

If you use llm-analysis in your work, please cite:

Cheng Li. (2023). LLM-Analysis: Latency and Memory Analysis of Transformer Models for Training and Inference. GitHub repository, https://github.com/cli99/llm-analysis.

or

@misc{llm-analysis-chengli,
  author = {Cheng Li},
  title = {LLM-Analysis: Latency and Memory Analysis of Transformer Models for Training and Inference},
  year = {2023},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/cli99/llm-analysis}},
}

Contributing

Contributions and suggestions are welcome.

llm-analysis uses pre-commit to ensure code formatting is consistent. For pull requests with code contribution, please install the pre-commit (pip install pre-commit) as well as the used hooks (pip install in the repo), and format the code (runs automatically before each git commit) before submitting the PR.

Useful Links

  1. Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism
  2. ZeRO: Memory Optimizations Toward Training Trillion Parameter Models
  3. Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM
  4. Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model
  5. Reducing Activation Recomputation in Large Transformer Models
  6. Training Compute-Optimal Large Language Models
  7. Efficiently Scaling Transformer Inference
  8. Training Compute-Optimal Large Language Models
  9. Understanding INT4 Quantization for Transformer Models: Latency Speedup, Composability, and Failure Cases
  10. A Comprehensive Study on Post-Training Quantization for Large Language Models