Fix DeepSpeed Zero 3 Saving (axolotl-ai-cloud#709)

* Update train.py

* add zero3 check

* chore: lint

---------

Co-authored-by: Wing Lian <wing.lian@gmail.com>

src/axolotl/train.py

@@ -12,6 +12,7 @@
 import transformers.modelcard
 from datasets import Dataset
 from optimum.bettertransformer import BetterTransformer
+from transformers.deepspeed import is_deepspeed_zero3_enabled
 
 from axolotl.common.cli import TrainerCliArgs
 from axolotl.logging_config import configure_logging
@@ -134,6 +135,22 @@ def terminate_handler(_, __, model):
     # only save on rank 0, otherwise it corrupts output on multi-GPU when multiple processes attempt to write the same file
     if cfg.fsdp:
         trainer.save_model(cfg.output_dir)
+    elif cfg.deepspeed and is_deepspeed_zero3_enabled():
+        # Copied over from: https://github.com/huggingface/accelerate/blob/5ae611118057232f441055f7ef9ba0b0f2b8d533/docs/source/usage_guides/deepspeed.md#saving-and-loading
+        trainer.accelerator.wait_for_everyone()
+        unwrapped_model = trainer.accelerator.unwrap_model(trainer.model_wrapped)
+
+        # Saves the whole/unpartitioned fp16 model when in ZeRO Stage-3 to the output directory if
+        # `stage3_gather_16bit_weights_on_model_save` is True in DeepSpeed Config file or
+        # `zero3_save_16bit_model` is True in DeepSpeed Plugin.
+        # For Zero Stages 1 and 2, models are saved as usual in the output directory.
+        # The model name saved is `pytorch_model.bin`
+        unwrapped_model.save_pretrained(
+            cfg.output_dir,
+            is_main_process=trainer.accelerator.is_main_process,
+            save_function=trainer.accelerator.save,
+            state_dict=trainer.accelerator.get_state_dict(trainer.model_wrapped),
+        )
     elif cfg.local_rank == 0:
         if cfg.flash_optimum:
             model = BetterTransformer.reverse(model)