Upload job.sh with huggingface_hub

#1
Files changed (1) hide show
  1. job.sh +1 -2
job.sh CHANGED
@@ -36,6 +36,7 @@ echo "PYTORCH_CUDA_ALLOC_CONF=${PYTORCH_CUDA_ALLOC_CONF}"
36
  echo "CKPT_LOCAL_DIR=${CKPT_LOCAL_DIR} TMPDIR=${TMPDIR}"
37
 
38
  ARGS=(
 
39
  --model-size "${MODEL_SIZE:-1.7b}"
40
  --dataset "${DATASET:-HuggingFaceFW/fineweb}"
41
  --dataset-config "${DATASET_CONFIG:-sample-10BT}"
@@ -53,7 +54,6 @@ ARGS=(
53
  --save-interval "${SAVE_INTERVAL:-100}"
54
  --weights-save-interval "${WEIGHTS_SAVE_INTERVAL:-0}"
55
  --ckpt-local-dir "${CKPT_LOCAL_DIR}"
56
- --bucket-sync-interval "${BUCKET_SYNC_INTERVAL:-1}"
57
  --log-interval "${LOG_INTERVAL:-10}"
58
  --dtype auto
59
  --tokenizer "${TOKENIZER:-meta-llama/Llama-3.2-1B}"
@@ -66,7 +66,6 @@ ARGS=(
66
  )
67
 
68
  if [ -n "${ASYNC_BUCKET_URI:-${BUCKET_URI:-}}" ] && [ "${ASYNC_BUCKET_UPLOAD:-1}" = "1" ]; then
69
- ARGS+=(--async-bucket-uri "${ASYNC_BUCKET_URI:-${BUCKET_URI}}")
70
  fi
71
 
72
  if [ -n "${INIT_FROM:-}" ]; then
 
36
  echo "CKPT_LOCAL_DIR=${CKPT_LOCAL_DIR} TMPDIR=${TMPDIR}"
37
 
38
  ARGS=(
39
+ --sft
40
  --model-size "${MODEL_SIZE:-1.7b}"
41
  --dataset "${DATASET:-HuggingFaceFW/fineweb}"
42
  --dataset-config "${DATASET_CONFIG:-sample-10BT}"
 
54
  --save-interval "${SAVE_INTERVAL:-100}"
55
  --weights-save-interval "${WEIGHTS_SAVE_INTERVAL:-0}"
56
  --ckpt-local-dir "${CKPT_LOCAL_DIR}"
 
57
  --log-interval "${LOG_INTERVAL:-10}"
58
  --dtype auto
59
  --tokenizer "${TOKENIZER:-meta-llama/Llama-3.2-1B}"
 
66
  )
67
 
68
  if [ -n "${ASYNC_BUCKET_URI:-${BUCKET_URI:-}}" ] && [ "${ASYNC_BUCKET_UPLOAD:-1}" = "1" ]; then
 
69
  fi
70
 
71
  if [ -n "${INIT_FROM:-}" ]; then