fmbench 2.0.11

Benchmark performance of **any Foundation Model (FM)** deployed on **any AWS Generative AI service**, be it **Amazon SageMaker**, **Amazon Bedrock**, **Amazon EKS**, or **Amazon EC2**. The FMs could be deployed on these platforms either directly through `FMbench`, or, if they are already deployed then also they could be benchmarked through the **Bring your own endpoint** mode supported by `FMBench`.

FMBench

Benchmark any Foundation Model (FM) on any AWS Generative AI service [Amazon SageMaker, Amazon Bedrock, Amazon EKS, Amazon EC2, or Bring your own endpoint.]

Amazon Bedrock | Amazon SageMaker | Amazon EKS | Amazon EC2

FMBench is a Python package for running performance benchmarks and accuracy for any Foundation Model (FM) deployed on any AWS Generative AI service, be it Amazon SageMaker, Amazon Bedrock, Amazon EKS, or Amazon EC2. The FMs could be deployed on these platforms either directly through FMbench, or, if they are already deployed then also they could be benchmarked through the Bring your own endpoint mode supported by FMBench.

Here are some salient features of FMBench:

1. Highly flexible: in that it allows for using any combinations of instance types (g5, p4d, p5, Inf2), inference containers (DeepSpeed, TensorRT, HuggingFace TGI and others) and parameters such as tensor parallelism, rolling batch etc. as long as those are supported by the underlying platform.

2. Benchmark any model: it can be used to be benchmark open-source models, third party models, and proprietary models trained by enterprises on their own data. Benchmarking includes both performance benchmaking and model evaluations (accuracy measurement given ground truth). 🚨 NEW: Model evaluations done by a Panel of LLM Evaluators added in release 2.0.0 🚨

3. Run anywhere: it can be run on any AWS platform where we can run Python, such as Amazon EC2, Amazon SageMaker, or even the AWS CloudShell. It is important to run this tool on an AWS platform so that internet round trip time does not get included in the end-to-end response time latency.

Intro Video

Determine the optimal price|performance serving stack for your generative AI workload

Use FMBench to benchmark an LLM on any AWS generative AI service for price and performance (inference latency, transactions/minute). Here is one of the plots generated by FMBench to help answer the price performance question for the Llama2-13b model when hosted on Amazon SageMaker (the instance types in the legend have been blurred out on purpose, you can find them in the actual plot generated on running FMBench).

Determine the optimal model for your generative AI workload

Use FMBench to determine model accuracy using a panel of LLM evaluators (PoLL [1]). Here is one of the plots generated by FMBench to help answer the accuracy question for various FMs on Amazon Bedrock (the model ids in the charts have been blurred out on purpose, you can find them in the actual plot generated on running FMBench).

Models benchmarked

Configuration files are available in the configs folder for the following models in this repo.

Llama3 on Amazon SageMaker

Llama3 is now available on SageMaker (read blog post), and you can now benchmark it using FMBench. Here are the config files for benchmarking Llama3-8b-instruct and Llama3-70b-instruct on ml.p4d.24xlarge, ml.inf2.24xlarge and ml.g5.12xlarge instances.

• Config file for Llama3-8b-instruct on ml.p4d.24xlarge and ml.g5.12xlarge.
• Config file for Llama3-70b-instruct on ml.p4d.24xlarge and ml.g5.48xlarge.
• Config file for Llama3-8b-instruct on ml.inf2.24xlarge and ml.g5.12xlarge.

Full list of benchmarked models

Model	EC2 g5	EC2 p4	EC2 p5	EC2 Inf2/Trn1	SageMaker g4dn/g5/p3	SageMaker Inf2/Trn1	SageMaker P4	SageMaker P5	Bedrock On-demand throughput	Bedrock provisioned throughput
Anthropic Claude-3 Sonnet									✅	✅
Anthropic Claude-3 Haiku									✅
Mistral-7b-instruct				✅	✅		✅	✅	✅
Mistral-7b-AWQ								✅
Mixtral-8x7b-instruct									✅
Llama3.2-1b instruct	✅
Llama3.2-3b instruct	✅
Llama3.1-8b instruct	✅	✅	✅	✅	✅	✅			✅
Llama3.1-70b instruct				✅		✅			✅
Llama3-8b instruct	✅			✅	✅	✅	✅	✅	✅
Llama3-70b instruct	✅				✅	✅	✅		✅
Llama2-13b chat					✅	✅	✅		✅
Llama2-70b chat					✅	✅	✅		✅
Amazon Titan text lite									✅
Amazon Titan text express									✅
Cohere Command text									✅
Cohere Command light text									✅
AI21 J2 Mid									✅
AI21 J2 Ultra									✅
Gemma-2b					✅
Phi-3-mini-4k-instruct					✅
distilbert-base-uncased					✅

New in this release

2.0.11

1. Llama3.2-1b and Llama3.2-3b support on EC2 g5.
2. Llama3-8b on EC2 g6e instances.

2.0.9

1. Triton-djl support for AWS Chips.
2. Tokenizer files are now downloaded directly from Hugging Face (unless provided manually as before)

2.0.8

1. Support Triton-TensorRT for GPU instances and Triton-vllm for AWS Chips.
2. Misc. bug fixes.

Release history

Getting started

FMBench is available as a Python package on PyPi and is run as a command line tool once it is installed. All data that includes metrics, reports and results are stored in an Amazon S3 bucket.

[!IMPORTANT] 💡 All documentation for FMBench is available on the FMBench website

You can run FMBench on either a SageMaker notebook or on an EC2 VM. Both options are described here as part of the documentation. You can even run FMBench as a Docker container A Quickstart guide for SageMaker is bring provided below as well.

👉 The following sections are discussing running FMBench the tool, as different from where the FM is actually deployed. For example, we could run FMBench on EC2 but the model being deployed is on SageMaker or even Bedrock.

Quickstart

FMBench on a SageMaker Notebook

1. Each FMBench run works with a configuration file that contains the information about the model, the deployment steps, and the tests to run. A typical FMBench workflow involves either directly using an already provided config file from the configs folder in the FMBench GitHub repo or editing an already provided config file as per your own requirements (say you want to try benchmarking on a different instance type, or a different inference container etc.).

   👉 A simple config file with key parameters annotated is included in this repo, see config-llama2-7b-g5-quick.yml. This file benchmarks performance of Llama2-7b on an ml.g5.xlarge instance and an ml.g5.2xlarge instance. You can use this config file as it is for this Quickstart.

2. Launch the AWS CloudFormation template included in this repository using one of the buttons from the table below. The CloudFormation template creates the following resources within your AWS account: Amazon S3 buckets, Amazon IAM role and an Amazon SageMaker Notebook with this repository cloned. A read S3 bucket is created which contains all the files (configuration files, datasets) required to run FMBench and a write S3 bucket is created which will hold the metrics and reports generated by FMBench. The CloudFormation stack takes about 5-minutes to create.

   AWS Region	Link
   us-east-1 (N. Virginia)
   us-west-2 (Oregon)
   us-gov-west-1 (GovCloud West)

3. Once the CloudFormation stack is created, navigate to SageMaker Notebooks and open the fmbench-notebook.

4. On the fmbench-notebook open a Terminal and run the following commands.

   conda create --name fmbench_python311 -y python=3.11 ipykernel
   source activate fmbench_python311;
   pip install -U fmbench
   

5. Now you are ready to fmbench with the following command line. We will use a sample config file placed in the S3 bucket by the CloudFormation stack for a quick first run.

   1. We benchmark performance for the Llama2-7b model on a ml.g5.xlarge and a ml.g5.2xlarge instance type, using the huggingface-pytorch-tgi-inference inference container. This test would take about 30 minutes to complete and cost about $0.20.

   2. It uses a simple relationship of 750 words equals 1000 tokens, to get a more accurate representation of token counts use the Llama2 tokenizer (instructions are provided in the next section). It is strongly recommended that for more accurate results on token throughput you use a tokenizer specific to the model you are testing rather than the default tokenizer. See instructions provided later in this document on how to use a custom tokenizer.

      account=`aws sts get-caller-identity | jq .Account | tr -d '"'`
      region=`aws configure get region`
      fmbench --config-file s3://sagemaker-fmbench-read-${region}-${account}/configs/llama2/7b/config-llama2-7b-g5-quick.yml > fmbench.log 2>&1
      

   3. Open another terminal window and do a tail -f on the fmbench.log file to see all the traces being generated at runtime.

      tail -f fmbench.log
      

   4. 👉 For streaming support on SageMaker and Bedrock checkout these config files:

      1. config-llama3-8b-g5-streaming.yml
      2. config-bedrock-llama3-streaming.yml

6. The generated reports and metrics are available in the sagemaker-fmbench-write-<replace_w_your_aws_region>-<replace_w_your_aws_account_id> bucket. The metrics and report files are also downloaded locally and in the results directory (created by FMBench) and the benchmarking report is available as a markdown file called report.md in the results directory. You can view the rendered Markdown report in the SageMaker notebook itself or download the metrics and report files to your machine for offline analysis.

If you would like to understand what is being done under the hood by the CloudFormation template, see the DIY version with gory details

FMBench on SageMaker in GovCloud

No special steps are required for running FMBench on GovCloud. The CloudFormation link for us-gov-west-1 has been provided in the section above.

1. Not all models available via Bedrock or other services may be available in GovCloud. The following commands show how to run FMBench to benchmark the Amazon Titan Text Express model in the GovCloud. See the Amazon Bedrock GovCloud page for more details.

account=`aws sts get-caller-identity | jq .Account | tr -d '"'`
region=`aws configure get region`
fmbench --config-file s3://sagemaker-fmbench-read-${region}-${account}/configs/bedrock/config-bedrock-titan-text-express.yml > fmbench.log 2>&1

Results

Depending upon the experiments in the config file, the FMBench run may take a few minutes to several hours. Once the run completes, you can find the report and metrics in the local results-* folder in the directory from where FMBench was run. The rpeort and metrics are also written to the write S3 bucket set in the config file.

Here is a screenshot of the report.md file generated by FMBench.

Benchmark models deployed on different AWS Generative AI services (Docs)

FMBench comes packaged with configuration files for benchmarking models on different AWS Generative AI services i.e. Bedrock, SageMaker, EKS and EC2 or bring your own endpoint even.

Enhancements

View the ISSUES on GitHub and add any you might think be an beneficial iteration to this benchmarking harness.

Security

See CONTRIBUTING for more information.

License

This library is licensed under the MIT-0 License. See the LICENSE file.

Star History

Support

• Schedule Demo 👋 - send us an email 🙂
• Community Discord 💭
• Our emails ✉️ [e-mail] / [e-mail]

Contributors

References

[1] Pat Verga et al., "Replacing Judges with Juries: Evaluating LLM Generations with a Panel of Diverse Models", arXiv:2404.18796, 2024.