checkpoint model on first step callback (#2906)

* checkpoint model on first step callback

* remove debug

* add test cases; update existing tests not to save on first step

* move test out of solo

* delete

* default to False

* typo

examples/jamba/qlora.yaml

@@ -54,3 +54,5 @@ evals_per_epoch:
 saves_per_epoch: 1
 weight_decay: 0.0
 special_tokens:
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/jamba/qlora_deepspeed.yaml

@@ -55,3 +55,5 @@ saves_per_epoch: 1
 deepspeed: deepspeed_configs/zero2.json
 weight_decay: 0.0
 special_tokens:
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/jamba/qlora_fsdp_large.yaml

@@ -64,3 +64,5 @@ fsdp_config:
   fsdp_transformer_layer_cls_to_wrap: JambaAttentionDecoderLayer,JambaMambaDecoderLayer
   fsdp_state_dict_type: FULL_STATE_DICT
   fsdp_sharding_strategy: FULL_SHARD
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config