rende-nlp-cp / code /cp_pod_bootstrap.sh

Upload code/cp_pod_bootstrap.sh with huggingface_hub

053459f verified 3 months ago

2.32 kB

#!/bin/bash
# Runs ON a RunPod pod. Generates one clone shard, uploads it to HF, self-terminates.
# Durability rule (gpu-runbook): the pod owns its own result delivery -- a Mac-side watcher
# can be reaped, so nothing may depend on one. Incremental saves + HF upload + self-kill.
set -x
SHARD=$1; N_SHARDS=$2; N_SEQS=$3; HARVEST=$4; THIN=$5; CHUNK=$6; ORDER=${7:-9}
case "$ORDER" in 3) NL=2;; 9) NL=4;; 33) NL=6;; *) echo "bad order"; exit 1;; esac
cd /root

apt-get update -qq && apt-get install -y -qq tmux git >/dev/null 2>&1
pip install -q huggingface_hub numpy

python - "$NL" <<'PY'
import os
from huggingface_hub import hf_hub_download
os.makedirs("/root/results/nlp_cp", exist_ok=True)
import sys
for f in [f"factored{sys.argv[1]}.pt", "train.npy", "test.npy"]:
    p = hf_hub_download("bgradowhite/rende-nlp-fig1", f"results/{f}",
                        repo_type="dataset", local_dir="/root/dl")
    os.replace(p, f"/root/results/nlp_cp/{f}")
print("data ready")
PY

# GPU health before investing (gpu-runbook non-negotiable)
nvidia-smi --query-gpu=name,memory.total,ecc.errors.uncorrected.volatile.total --format=csv
python -c "import torch;assert torch.cuda.is_available();print('cuda OK',torch.cuda.get_device_name(0))" || {
  echo "GPU UNUSABLE - self-terminating"; runpodctl remove pod "$RUNPOD_POD_ID"; exit 1; }

tmux new-session -d -s clone "cd /root && PYTHONUNBUFFERED=1 python -u nlp_cp_sweep.py clones \
  --shard $SHARD --n-shards $N_SHARDS --n-seqs $N_SEQS --harvest $HARVEST --thin $THIN \
  --chunk $CHUNK --order $ORDER --outdir /root/results/nlp_cp 2>&1 | tee /root/shard$SHARD.log; \
  echo EXIT=\$? >> /root/shard$SHARD.log"

while tmux has-session -t clone 2>/dev/null; do sleep 60; done

python - <<PY
from huggingface_hub import HfApi
api = HfApi(); api.create_repo("bgradowhite/rende-nlp-cp", repo_type="dataset", exist_ok=True)
for f, n in [("/root/results/nlp_cp/clone${ORDER}_shard$SHARD.npy", "clone${ORDER}_shard$SHARD.npy"),
             ("/root/shard$SHARD.log", "logs/shard$SHARD.log")]:
    try:
        api.upload_file(path_or_fileobj=f, path_in_repo=n,
                        repo_id="bgradowhite/rende-nlp-cp", repo_type="dataset")
        print("uploaded", n)
    except Exception as e:
        print("UPLOAD FAILED", n, e)
PY

runpodctl remove pod "$RUNPOD_POD_ID"

	#!/bin/bash
	# Runs ON a RunPod pod. Generates one clone shard, uploads it to HF, self-terminates.
	# Durability rule (gpu-runbook): the pod owns its own result delivery -- a Mac-side watcher
	# can be reaped, so nothing may depend on one. Incremental saves + HF upload + self-kill.
	set -x
	SHARD=$1; N_SHARDS=$2; N_SEQS=$3; HARVEST=$4; THIN=$5; CHUNK=$6; ORDER=${7:-9}
	case "$ORDER" in 3) NL=2;; 9) NL=4;; 33) NL=6;; *) echo "bad order"; exit 1;; esac
	cd /root

	apt-get update -qq && apt-get install -y -qq tmux git >/dev/null 2>&1
	pip install -q huggingface_hub numpy

	python - "$NL" <<'PY'
	import os
	from huggingface_hub import hf_hub_download
	os.makedirs("/root/results/nlp_cp", exist_ok=True)
	import sys
	for f in [f"factored{sys.argv[1]}.pt", "train.npy", "test.npy"]:
	p = hf_hub_download("bgradowhite/rende-nlp-fig1", f"results/{f}",
	repo_type="dataset", local_dir="/root/dl")
	os.replace(p, f"/root/results/nlp_cp/{f}")
	print("data ready")
	PY

	# GPU health before investing (gpu-runbook non-negotiable)
	nvidia-smi --query-gpu=name,memory.total,ecc.errors.uncorrected.volatile.total --format=csv
	python -c "import torch;assert torch.cuda.is_available();print('cuda OK',torch.cuda.get_device_name(0))" \|\| {
	echo "GPU UNUSABLE - self-terminating"; runpodctl remove pod "$RUNPOD_POD_ID"; exit 1; }

	tmux new-session -d -s clone "cd /root && PYTHONUNBUFFERED=1 python -u nlp_cp_sweep.py clones \
	--shard $SHARD --n-shards $N_SHARDS --n-seqs $N_SEQS --harvest $HARVEST --thin $THIN \
	--chunk $CHUNK --order $ORDER --outdir /root/results/nlp_cp 2>&1 \| tee /root/shard$SHARD.log; \
	echo EXIT=\$? >> /root/shard$SHARD.log"

	while tmux has-session -t clone 2>/dev/null; do sleep 60; done

	python - <<PY
	from huggingface_hub import HfApi
	api = HfApi(); api.create_repo("bgradowhite/rende-nlp-cp", repo_type="dataset", exist_ok=True)
	for f, n in [("/root/results/nlp_cp/clone${ORDER}_shard$SHARD.npy", "clone${ORDER}_shard$SHARD.npy"),
	("/root/shard$SHARD.log", "logs/shard$SHARD.log")]:
	try:
	api.upload_file(path_or_fileobj=f, path_in_repo=n,
	repo_id="bgradowhite/rende-nlp-cp", repo_type="dataset")
	print("uploaded", n)
	except Exception as e:
	print("UPLOAD FAILED", n, e)
	PY

	runpodctl remove pod "$RUNPOD_POD_ID"