feat: Change vllm frac to 0.6 (#31)

Signed-off-by: Parth Chadha <pchadha@nvidia.com>

Author: parthchadha

examples/configs/grpo_math_1B.yaml

@@ -59,7 +59,7 @@ policy:
     top_k: null
     vllm_cfg:
       tensor_parallel_size: 1
-      gpu_memory_utilization: 0.7
+      gpu_memory_utilization: 0.6
       max_model_len: ${policy.max_total_sequence_length}
 
 data: