Skip to content
Snippets Groups Projects

Draft: Resolve "Konfigurieren für llama3:70b und slrum"

2 files
+ 9
6
Compare changes
  • Side-by-side
  • Inline
Files
2
  • 7d99559a
    Changed minor things in scripts · 7d99559a
    Erik Jonas Hartnick authored
    - informative part at the start
    - CUDA_VISIBLE devices now exported
    - vllm proper context length now also on download, shouldn't matter much since on execution, it was correct already
+ 5
4
@@ -38,9 +38,10 @@ if [ ! -f $2 ]; then
fi
# Useful debug Information
srun echo "Tested: <running CHESS on SDS-BIRD>" # place a little reminder on what was tested here.
srun hostname
srun nvidia-smi
srun --ntasks=1 echo "Tested: <running CHESS on SDS-BIRD>" # place a little reminder on what was tested here.
srun --ntasks=1 hostname
srun --ntasks=1 nvidia-smi
srun --ntasks=1 echo ${CUDA_VISIBLE_DEVICES}
# Create user directory on /zpool1/slurm_data/ if it doesn't exist yet
srun --ntasks=1 $(if [ -d /zpool1/slurm_data/${KUERZEL} ]; then rm -r /zpool1/slurm_data/${KUERZEL}/*; else mkdir /zpool1/slurm_data/${KUERZEL}; fi)
@@ -127,7 +128,7 @@ source "${VENVPATH}/bin/activate" ;
srun --ntasks=1 pip3.11 install -r "${RECHESSROOT}/CHESS/requirements.txt"
# Load nl2sql model via vllm
srun --time='00:05:00' --ntasks=1 ${PYTHONVER} -m vllm.entrypoints.openai.api_server --model "AI4DS/NL2SQL_DeepSeek_33B" --load-format safetensors --dtype bfloat16 --max-model-len 16384 --download-dir "${RECHESSROOT}/vllm"
srun --time='00:05:00' --ntasks=1 ${PYTHONVER} -m vllm.entrypoints.openai.api_server --model "AI4DS/NL2SQL_DeepSeek_33B" --load-format safetensors --dtype bfloat16 --max-model-len 8192 --download-dir "${RECHESSROOT}/vllm"
srun --ntasks=1 ls -la "${VENVPATH}/lib/${PYTHONVER}/site-packages"
deactivate
# srun which pip3.11 ; # for debugging purposes
Loading