support AGENT_MODEL env var

scripts/hf_train_h200_entry.sh

@@ -9,6 +9,7 @@ STEPS="${STEPS:-300}"
 BATCH="${BATCH:-8}"
 NUM_GENS="${NUM_GENS:-4}"
 PER_TYPE="${PER_TYPE:-4}"
+AGENT_MODEL="${AGENT_MODEL:-Qwen/Qwen2.5-1.5B-Instruct}"
 MODEL_REPO="${HF_USERNAME}/promptops-arena-agent"
 
 echo "[train-h200] HF_USERNAME=${HF_USERNAME} STEPS=${STEPS} BATCH=${BATCH} NUM_GENS=${NUM_GENS}"
@@ -40,8 +41,9 @@ export TOKENIZERS_PARALLELISM=false
 
 mkdir -p outputs results
 
-echo "[train-h200] launching GRPO training"
+echo "[train-h200] launching GRPO training (model=${AGENT_MODEL})"
 python scripts/train_grpo.py \
+    --model "${AGENT_MODEL}" \
     --steps "${STEPS}" \
     --batch "${BATCH}" \
     --num-generations "${NUM_GENS}" \