ADLR/megatron-lm!1732 - New scaling figures on H100 GPUs

README.md

@@ -63,18 +63,18 @@ Megatron-Core can be used with [NVIDIA NeMo](https://www.nvidia.com/en-us/ai-dat
 
 
 # Training Speed and Scalability
-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 specific 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. Each cluster node has 8 NVIDIA 80GB A100 GPUs. The graph below shows that we scale nearly linear up to 1 trillion parameter models running on 3072 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.
+Our codebase is capable of efficiently training large language models (i.e., models with hundreds of billions of parameters) with both model and data parallelism. To demonstrate how our software scales with multiple GPUs and model sizes, we consider GPT models ranging from 2 billion parameters to 462 billion parameters. All models use a vocabulary size of 131,072 and a sequence length of 4096. We vary hidden size, number of attention heads, and number of layers to arrive at a specific model size. As the model size increases, we also modestly increase batch size. Our experiments use up to 6144 [H100](https://www.nvidia.com/en-us/data-center/h100/) GPUs. We perform fine-grained overlapping of data-parallel (`--overlap-grad-reduce --overlap-param-gather`), tensor-parallel (`--tp-comm-overlap`) and pipeline-parallel communication (enabled by default) with computation to improve scalability. The reported throughputs are measured for end-to-end training and include all operations including data loading, optimizer steps, communication, and even logging. Note that we did not train these models to convergence.
 
-![Scaling Graph](images/Achieved_petaFLOPs.png)
+![Model table](images/model_table.png)
 
-The following table shows both model (MFU) and hardware (HFU) FLOPs utilization for select configurations up to 1T parameters (see [our paper](https://arxiv.org/pdf/2205.05198) for a description of how these are calculated). As the model size increases, we achieve better GPU utilization. For the one trillion parameter model, we reach a MFU and HFU of 56.3% and 57.0%, respectively. Note that these numbers are also measured on benchmark runs and in this case are measured using a data parallel size of one. Data parallelism introduces some overhead due to the gradient all-reduce required between the data parallel groups. However, for large transformer models, this overhead is not large and can almost entirely eliminated by overlapping the gradient all-reduce with backpropagation.
+Our weak scaled results show superlinear scaling (MFU increases from 41% for the smallest model considered to 47-48% for the largest models); this is because larger GEMMs have higher arithmetic intensity and are consequently more efficient to execute.
+
+![Weak scaling](images/weak_scaling.png)
+
+We also strong scaled the standard GPT-3 model (our version has slightly more than 175 billion parameters due to larger vocabulary size) from 96 H100 GPUs to 4608 GPUs, using the same batch size of 1152 sequences throughout. Communication becomes more exposed at larger scale, leading to a reduction in MFU from 47% to 42%.
+
+![Strong scaling](images/strong_scaling.png)
 
-| Model Size | Model FLOPs Utilization | Hardware FLOPs Utilization |
-| :---: | :---: | :---: |
-| 22B   | 41.5% | 43.7% |
-| 175B  | 51.4% | 52.8% |
-| 530B  | 56.0% | 57.0% |
-| 1T    | 56.3% | 57.0% |
 
 # Setup
 We strongly recommend using the latest release of [NGC's PyTorch container](https://ngc.nvidia.com/catalog/containers/nvidia:pytorch) with DGX nodes. If you can't use this for some reason, use the latest pytorch, cuda, nccl, and NVIDIA [APEX](https://github.com/NVIDIA/apex#quick-start) releases.  Data preprocessing requires [NLTK](https://www.nltk.org/install.html), though this is not required for training, evaluation, or downstream tasks.