fast local checkpoints, fewer bucket writes
Browse files
job.sh
CHANGED
|
@@ -24,9 +24,11 @@ ARGS=(
|
|
| 24 |
--out-dir "${OUT_DIR}"
|
| 25 |
--hub-repo-id "${HUB_REPO_ID:-ProjectScugnizz/scugnizz-llama-pcs}"
|
| 26 |
--hub-path "${HUB_PATH:-training-runs/pretrain-fineweb-llama-pcs}"
|
| 27 |
-
--eval-interval "${EVAL_INTERVAL:-
|
| 28 |
-
--save-interval "${SAVE_INTERVAL:-
|
| 29 |
-
--weights-save-interval "${WEIGHTS_SAVE_INTERVAL:-
|
|
|
|
|
|
|
| 30 |
--log-interval "${LOG_INTERVAL:-10}"
|
| 31 |
--dtype auto
|
| 32 |
--tokenizer "${TOKENIZER:-meta-llama/Llama-3.2-1B}"
|
|
|
|
| 24 |
--out-dir "${OUT_DIR}"
|
| 25 |
--hub-repo-id "${HUB_REPO_ID:-ProjectScugnizz/scugnizz-llama-pcs}"
|
| 26 |
--hub-path "${HUB_PATH:-training-runs/pretrain-fineweb-llama-pcs}"
|
| 27 |
+
--eval-interval "${EVAL_INTERVAL:-100}"
|
| 28 |
+
--save-interval "${SAVE_INTERVAL:-100}"
|
| 29 |
+
--weights-save-interval "${WEIGHTS_SAVE_INTERVAL:-0}"
|
| 30 |
+
--ckpt-local-dir "${CKPT_LOCAL_DIR:-/tmp/scugnizz-ckpts}"
|
| 31 |
+
--bucket-sync-interval "${BUCKET_SYNC_INTERVAL:-1}"
|
| 32 |
--log-interval "${LOG_INTERVAL:-10}"
|
| 33 |
--dtype auto
|
| 34 |
--tokenizer "${TOKENIZER:-meta-llama/Llama-3.2-1B}"
|