No slurm (#39)

* changed metaseq_internal to metaseq

* fixed naming of script in docs, added resharding script without slurm

metaseq/scripts/reshard_mp_launch_no_slurm.sh

@@ -0,0 +1,26 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates. All Rights Reserved.
+#
+# This source code is licensed under the MIT license found in the
+# LICENSE file in the root directory of this source tree.
+
+prefix=$1
+save_dir=$2
+mparts=$3
+tgt_size=$4
+shift 4
+mkdir -p $save_dir
+let "last_part=$mparts -1"
+echo "$@"
+for i in $(seq 0 $last_part)
+do
+
+  echo "python -m metaseq.scripts.reshard_mp $prefix $save_dir --part $i --target-ddp-size $tgt_size"
+  jname=reshard_mp"$i"_ddp"$tgt_size"
+  echo $jname
+  python3 -m metaseq.scripts.reshard_mp $prefix $save_dir --part $i --target-ddp-size $tgt_size &
+done
+echo "Waiting on jobs..."
+wait $(jobs -p)
+echo "Done"
+

projects/OPT/download_opt175b.md

@@ -17,7 +17,11 @@ bash metaseq/scripts/download_opt175b.sh "<presigned_url_given_in_email>"
 ## Reshard the shards
 To consolidate the 992 shards into 8 files model-parallel evaluation, run (assuming you have SLURM set up already):
 ```
-bash metaseq/scripts/reshard_sbatch.sh <directory_where_all_the_shards_are>/checkpoint_last <output_dir>/ 8 1
+bash metaseq/scripts/reshard_mp_launch.sh <directory_where_all_the_shards_are>/checkpoint_last <output_dir>/ 8 1
+```
+If you don't have slurm on your machine, run:
+```
+bash metaseq/scripts/reshard_mp_launch_no_slurm.sh <directory_where_all_the_shards_are>/checkpoint_last <output_dir>/ 8 1
 ```
 
 Note that most of our models expect to run with Model (Tensor) Parallelism. For smaller models, some