minor fix

docs/checkpoint.md

@@ -75,5 +75,5 @@ A seed checkpoint does initialization of the model on a single CPU, and can be l
 To create a seed checkpoint, use the same model config as you use for training.
 e.g.
 ```bash
-NGPU=1 CONFIG=<path_to_model_config> ./run_llama_train.sh --checkpoint.enable_checkpoint --checkpoint.create_seed_checkpoint --training.data_parallel_shard_degree 1
+NGPU=1 CONFIG=<path_to_model_config> ./run_llama_train.sh --checkpoint.enable_checkpoint --checkpoint.create_seed_checkpoint --training.data_parallel_replicate_degree 1 --training.data_parallel_shard_degree 1 --training.tensor_parallel_degree 1 --experimental.pipeline_parallel_degree 1 --experimental.context_parallel_degree 1
 ```

torchtitan/parallelisms/pipelining_utils.py

@@ -107,7 +107,7 @@ def build_pipeline_schedule(job_config, stages, loss_fn):
     )
     logger.info(
         f"Using pipeline schedule {job_config.experimental.pipeline_parallel_schedule} \
-with {n_microbatches} and {num_total_stages} stages."
+with {n_microbatches} microbatches and {num_total_stages} stages."
     )
 
     if pp_schedule_csv:

train.py

@@ -201,7 +201,10 @@ def loss_fn(pred, labels):
     if job_config.checkpoint.create_seed_checkpoint:
         assert (
             world_size == 1
-        ), "Must create seed-checkpoint using one gpu, to disable sharding"
+        ), "Must create seed checkpoint using a single device, to disable sharding"
+        assert (
+            job_config.checkpoint.enable_checkpoint
+        ), "Must enable checkpointing when creating a seed checkpoint"
         checkpoint.save(curr_step=0, force=True)
         logger.info("Created seed checkpoint")
         return