@@ -11,7 +11,7 @@ Below are some of the projects where we have directly used Megatron:
*[Scaling Language Model Training to a Trillion Parameters Using Megatron](https://arxiv.org/pdf/2104.04473.pdf)
*[Training Question Answering Models From Synthetic Data](https://www.aclweb.org/anthology/2020.emnlp-main.468.pdf)
Our codebase is capable of efficiently training very large (hundreds of billions of parameters) language models with both model and data parallelism. To demonstrate how the code scales with multiple GPUs and model sizes, we consider GPT models from 1 billion all the way to 1 trillion parameters. All models use a vocabulary size of 51,200 and a sequence length of 2048. We vary hidden size, number of attention heads, and number of layers to arrive at a specifc model size. As the model size increases, we also modestly increase the batch size. We leverage [NVIDIA's Selene supercomputer](https://www.top500.org/system/179842/) to perform scaling studies and use up to 3072 [A100](https://www.nvidia.com/en-us/data-center/a100/) GPUs for the largest model. The table below shows the model configurations along with the achieved FLOPs (both per GPU and aggregate over all GPUs). Note that the FLOPs are measured for end-to-end training, i.e., includes all operations including data loading, optimization, and even logging.
Our codebase is capable of efficiently training very large (hundreds of billions of parameters) language models with both model and data parallelism. To demonstrate how the code scales with multiple GPUs and model sizes, we consider GPT models from 1 billion all the way to 1 trillion parameters. All models use a vocabulary size of 51,200 and a sequence length of 2048. We vary hidden size, number of attention heads, and number of layers to arrive at a specifc model size. As the model size increases, we also modestly increase the batch size. We leverage [NVIDIA's Selene supercomputer](https://www.top500.org/system/179842/) to perform scaling studies and use up to 3072 [A100](https://www.nvidia.com/en-us/data-center/a100/) GPUs for the largest model. The table below shows the model configurations along with the achieved FLOPs (both per GPU and aggregate over all GPUs). Note that these results are from benchmark runs and these models were not trained to convergence; however, the FLOPs are measured for end-to-end training, i.e., includes all operations including data loading, optimization, and even logging.
...
...
@@ -48,13 +48,6 @@ We have tested Megatron with [NGC's PyTorch container](https://ngc.nvidia.com/ca
To use this repository, please install the latest supported versions of PyTorch with GPU support (python 3.8, pytorch 1.8, cuda 11.1, and nccl 2.8.3 and above) and NVIDIA [APEX](https://github.com/NVIDIA/apex#quick-start). We strongly recommend using one of [NGC's recent PyTorch containers](https://ngc.nvidia.com/catalog/containers/nvidia:pytorch)(the latest compatible version at time of publication can be pulled with `docker pull nvcr.io/nvidia/pytorch:20.12-py3`). Data preprocessing requires [NLTK](https://www.nltk.org/install.html), though this is not required for training, evaluation, or downstream tasks.
<!--
To use megatron you can either clone the repo or install it via pip (make sure python3-dev is installed):
<pre>
pip install megatron-lm
</pre>
-->
## Downloading Checkpoints
We have provided pretrained [BERT-345M](https://ngc.nvidia.com/catalog/models/nvidia:megatron_bert_345m) and [GPT-345M](https://ngc.nvidia.com/catalog/models/nvidia:megatron_lm_345m) checkpoints for use to evaluate or finetuning downstream tasks. To access these checkpoints, first [sign up](https://ngc.nvidia.com/signup) for and [setup](https://ngc.nvidia.com/setup/installers/cli) the NVIDIA GPU Cloud (NGC) Registry CLI. Further documentation for downloading models can be found in the [NGC documentation](https://docs.nvidia.com/dgx/ngc-registry-cli-user-guide/index.html#topic_6_4_1).
Several downstream tasks are described for both GPT and BERT models below. They can be run in distributed and model parallel modes with the same changes used in the training scripts.
## GPT Text Generation
`bash examples/generate_text.sh`
We generate text samples using largely the GPT pretraining script. Few changes need to make, such as we need to provide the path to the pretrained checkpoint, the length of the output samples, whether to generate texts unconditionally (`--num-samples` to denote how many samples to generate) or conditional (need to pass `--sample-input-file <filename>` where each line of the file will be used as the conditional texts). There are few optional parameters to play, e.g. `top-k`, `top-p`, or `greedy` (set top-k and top-p to 0) sampling..
We have included a simple REST server to use for text generation in `tools/run_text_generation_server.py`. You run it much like you would start a pretraining job, specifying an appropriate pretrained checkpoint. There are also few optional parameters: `temperature`, `top-k`and `top-p`. See `--help` or the source file for more information. See [examples/run_text_generation_server_345M.sh](examples/run_text_generation_server_345M.sh) for an example of how to run the server.
Once the server is running you can use `tools/text_generation_cli.py` to query it, it takes one argument which is the host the server is running on.
<pre>
CHECKPOINT_PATH=checkpoints/gpt2_345m
VOCAB_FILE=gpt2-vocab.json
MERGE_FILE=gpt2-merges.txt
GPT_ARGS=<same as those in <ahref="#gpt-pretraining">GPT pretraining</a> above>
tools/text_generation_cli.py localhost
</pre>
MAX_OUTPUT_SEQUENCE_LENGTH=1024
TEMPERATURE=1.0
TOP_P=0.9
NUMBER_OF_SAMPLES=2
OUTPUT_FILE=samples.json
You can also use CURL or any other tools to query the server directly:
