38 lines
1.8 KiB
Bash
Executable File
38 lines
1.8 KiB
Bash
Executable File
#!/bin/bash
|
|
# vast.ai node bootstrap for jspace (see vast-ai-notes.md for pitfalls)
|
|
set -x
|
|
source /venv/main/bin/activate
|
|
uv pip install -q "transformers==5.13.*" datasets accelerate hf_transfer
|
|
export HF_HOME=/dev/shm/hf; mkdir -p /dev/shm/hf
|
|
cd /dev/shm && git clone https://git.draic.info/nils/jspace.git 2>/dev/null || (cd jspace && git pull origin main)
|
|
cd /dev/shm/jspace
|
|
rclone copy jspace:jspace/results-loop /dev/shm/jspace/results-loop/ --include "*.json" --include "adapter_code_e399.pt"
|
|
export HF_HUB_ENABLE_HF_TRANSFER=1
|
|
# downloads can hang mid-transfer on some vast.ai networks (xet stall):
|
|
# retry loop, resume is free
|
|
# PIN revisions: upstream updated gemma-4-12B-it's chat template (added a
|
|
# thought channel) mid-project; unpinned downloads silently change behavior.
|
|
for spec in "google/gemma-4-12B-it 0e2b1058541244490925fbacf8972041435691ac" \
|
|
"google/gemma-4-E2B-it 9dbdf8a839e4e9e0eb56ed80cc8886661d3817cf"; do
|
|
set -- $spec
|
|
for i in 1 2 3 4 5; do
|
|
timeout 900 hf download "$1" --revision "$2" && break
|
|
echo "RETRY $i: $1"; sleep 5
|
|
done
|
|
echo "$2" > "$HF_HOME/hub/models--${1//\//--}/refs/main"
|
|
done
|
|
# probe gate: one verified generation before anything batch
|
|
export HF_HUB_OFFLINE=1 JLENS_MODEL=google/gemma-4-12B-it
|
|
python - <<'PY'
|
|
import torch, sys
|
|
sys.path.insert(0, "/dev/shm/jspace/scripts"); sys.path.insert(0, "/dev/shm/jspace")
|
|
from jlens.core import load_model
|
|
m, tok = load_model(dtype=torch.bfloat16)
|
|
ids = tok.apply_chat_template([{"role":"user","content":"What is 2+2? Just the number."}],
|
|
return_tensors="pt", add_generation_prompt=True, return_dict=True)["input_ids"].cuda()
|
|
out = m.generate(ids, max_new_tokens=8, do_sample=False)
|
|
txt = tok.decode(out[0, ids.shape[1]:], skip_special_tokens=True)
|
|
print("PROBE:", repr(txt)); assert "4" in txt, "PROBE FAILED"
|
|
PY
|
|
echo NODE-READY
|