Upload Evaluation/setup_and_run.sh with huggingface_hub

Evaluation/setup_and_run.sh

@@ -25,15 +25,24 @@
 
 set -e
 
-MODEL_PATH="${1:?Usage: bash Evaluation/setup_and_run.sh /path/to/model_dir [tokenizer_path] [tasks]}"
+MODEL_PATH="${1:?Usage: bash Evaluation/setup_and_run.sh /path/to/model_dir [tokenizer_path] [tasks] [batch_size]}"
 TOKENIZER="${2:-Base/checkpoints/EleutherAI/pythia-160m}"
 TASKS="${3:-hellaswag,arc_easy,arc_challenge,piqa,winogrande,openbookqa,boolq,lambada_openai}"
+BATCH_SIZE="${4:-0}"
+
+# Resource safety knobs (can be overridden via env)
+RESERVE_CORES="${RESERVE_CORES:-2}"
+CPU_THREAD_FRACTION="${CPU_THREAD_FRACTION:-0.85}"
+BATCH_HEADROOM="${BATCH_HEADROOM:-0.80}"
 
 echo "=========================================="
 echo "  LUNA — Benchmark Runner"
 echo "  Model     : $MODEL_PATH"
 echo "  Tokenizer : $TOKENIZER"
 echo "  Tasks     : $TASKS"
+echo "  Batch     : $BATCH_SIZE (<=0 means auto)"
+echo "  CPU tune  : reserve=$RESERVE_CORES, fraction=$CPU_THREAD_FRACTION"
+echo "  GPU tune  : batch_headroom=$BATCH_HEADROOM"
 echo "=========================================="
 
 # ── 1. Install dependencies ───────────────────────────────────────────────────
@@ -59,7 +68,11 @@ python Evaluation/run_benchmarks.py \
     --model_path "$MODEL_PATH" \
     --tokenizer_path "$TOKENIZER" \
     --tasks "$TASKS" \
-    --batch_size 16
+    --batch_size "$BATCH_SIZE" \
+    --auto_resources true \
+    --reserve_cores "$RESERVE_CORES" \
+    --cpu_thread_fraction "$CPU_THREAD_FRACTION" \
+    --batch_headroom "$BATCH_HEADROOM"
 
 echo ""
 echo "=========================================="