@@ -288,7 +288,7 @@ Even when we set the batch size to 1 and use gradient accumulation we can still
Gradient checkpointing strikes a compromise between the two approaches and saves strategically selected activations throughout the computational graph so only a fraction of the activations need to be re-computed for the gradients. See [this great article](https://medium.com/tensorflow/fitting-larger-networks-into-memory-583e3c758ff9) explaining the ideas behind gradient checkpointing.
To enable gradient checkpointing in the [`Trainer`] we only need ot pass it as a flag to the [`TrainingArguments`]. Everything else is handled under the hood:
To enable gradient checkpointing in the [`Trainer`] we only need to pass it as a flag to the [`TrainingArguments`]. Everything else is handled under the hood:
For example, if you have [NVIDIA/apex](https://github.com/NVIDIA/apex) installed `--optim adamw_apex_fused` will give you the fastest training experience among all supported AdamW optimizers.
On the other hand [8bit BNB optimizer](https://github.com/facebookresearch/bitsandbytes) can save 3/4 of memory normally used by a typical AdamW optimizer if it is configured to quantize all optimizer states, but in some situations only some optimizer states are quintized and then more memory is used. XXX: update once https://github.com/huggingface/transformers/pull/15622 is merged.
On the other hand [8bit BNB optimizer](https://github.com/facebookresearch/bitsandbytes) can save 3/4 of memory normally used by a typical AdamW optimizer if it is configured to quantize all optimizer states, but in some situations only some optimizer states are quintized and then more memory is used.
Let's get a feel for the numbers and use for example use a 3B-parameter model, like `t5-3b`. Note that since a Gigabyte correpsonds to a billion bytes we can simply multiply the parameters (in billions) with the number of necessary bytes per parameter to get Gigabytes of GPU memory usage:
