The interleaved pipelining schedule (more details in Section 2.2.2 of [our paper](https://arxiv.org/pdf/2104.04473.pdf)) can be enabled using the `--num-layers-per-virtual-pipeline-stage` argument, which controls the number of transformer layers in a virtual stage (by default with the non-interleaved schedule, each GPU will execute a single virtual stage with `NUM_LAYERS / PIPELINE_MP_SIZE` transformer layers). The total number of layers in the transformer model should be divisible by this argument value. Additionally, the number of microbatches in the pipeline (computed as `GLOBAL_BATCH_SIZE / (DATA_PARALLEL_SIZE * MICRO_BATCH_SIZE)`) should be divisible by the `PIPELINE_MP_SIZE` when using this schedule (this condition is checked in an assertion in the code). The interleaved schedule is not supported for pipelines with 2 stages (`PIPELINE_MP_SIZE=2`).
The interleaved pipelining schedule (more details in Section 2.2.2 of [our paper](https://arxiv.org/pdf/2104.04473.pdf)) can be enabled using the `--num-layers-per-virtual-pipeline-stage` argument, which controls the number of transformer layers in a virtual stage (by default with the non-interleaved schedule, each GPU will execute a single virtual stage with `NUM_LAYERS / PIPELINE_MP_SIZE` transformer layers). The total number of layers in the transformer model should be divisible by this argument value. Additionally, the number of microbatches in the pipeline (computed as `GLOBAL_BATCH_SIZE / (DATA_PARALLEL_SIZE * MICRO_BATCH_SIZE)`) should be divisible by the `PIPELINE_MP_SIZE` when using this schedule (this condition is checked in an assertion in the code). The interleaved schedule is not supported for pipelines with 2 stages (`PIPELINE_MP_SIZE=2`).
## Activation Checkpointing and Recomputation
To reduce GPU memory usage so deploy a large model to a training system, we support activation checkpointing and recomputation. We use a Transformer layer as the unit of checkpointing because the activation size bloats in the middle of a Transformer layer so checkpointing the input of a Transformer layer is storage-efficient. We support two activation checkpointing methods: `uniform` and `block`.
Uniform method uniformly divides the Transformer layers into groups of layers and stores the input activations of each group in the memory. The baseline group size is 1 and, in this case, the input activation of each Transformer layer is checkpointed. When the GPU memory is insufficient, increasing the number of layers per group reduces the memory usage thus enables running a bigger model. For example, when using the number of layers per group of 4, the input activation of each group of 4 Transformer layers is checkpointed.
Block method checkpoints the input activations of a set number of individual Transformer layers per pipeline stage and do the rest of layers without any checkpointing. This method can be used to skip checkpointing some Transformer layers until the GPU memory is fully used, which is applicable only when there is unused GPU memory. Checkpointing fewer transformer layers avoids unnecessary activation recomputation in the backprop thus improves training performance. For example, when we specify 5 layers to checkpoint of 8 layers per pipeline stage, the input activations of only the first 5 Transformer layers are checkpointed and activation recomputation for the rest 3 layers is not needed in the backprop.
## GPT-3 Example
## GPT-3 Example
In `examples/pretrain_gpt3_175B.sh` we have provided an example of how to configure Megatron to run [GPT-3](https://arxiv.org/abs/2005.14165) with 175 billion parameters on 1024 GPUs. The script is designed for [slurm](https://slurm.schedmd.com/documentation.html) with [pyxis](https://github.com/NVIDIA/pyxis) plugin but can be easily adopted to any other scheduler. It uses 8-way and 16-way tensor and pipeline parallelism, respectively. With options `global-batch-size 1536` and `rampup-batch-size 16 16 5859375`, the training will start with global batch size 16 and linearly increase the global batch size to 1536 over 5,859,375 samples with incrmeental steps 16. The training dataset can be either a single set or a multiple datasets combined with a set of weights.
In `examples/pretrain_gpt3_175B.sh` we have provided an example of how to configure Megatron to run [GPT-3](https://arxiv.org/abs/2005.14165) with 175 billion parameters on 1024 GPUs. The script is designed for [slurm](https://slurm.schedmd.com/documentation.html) with [pyxis](https://github.com/NVIDIA/pyxis) plugin but can be easily adopted to any other scheduler. It uses 8-way and 16-way tensor and pipeline parallelism, respectively. With options `global-batch-size 1536` and `rampup-batch-size 16 16 5859375`, the training will start with global batch size 16 and linearly increase the global batch size to 1536 over 5,859,375 samples with incrmeental steps 16. The training dataset can be either a single set or a multiple datasets combined with a set of weights.
