Add 30B fp8 config

Signed-off-by: Guyue Huang <guyueh@nvidia.com>

Author: guyueh1

examples/configs/grpo_math_qwen30ba3b_megatron_fp8.yaml

@@ -0,0 +1,29 @@
+# GRPO Algorithm Configuration
+defaults: "grpo_math_qwen30ba3b_megatron.yaml"
+
+loss_fn:
+  use_importance_sampling_correction: true
+
+policy:
+  megatron_cfg:
+    moe_router_dtype: fp32
+    
+    fp8_cfg:
+      enabled: true
+      fp8: "e4m3"
+      fp8_recipe: "blockwise"
+      fp8_param: false
+
+    optimizer:
+      use_precision_aware_optimizer: false
+    
+    env_vars:
+      NVTE_FP8_BLOCK_SCALING_FP32_SCALES: "1"
+
+  generation:
+    vllm_cfg:
+      tensor_parallel_size: 8
+      expert_parallel_size: 8 # need to make moe_intermediate_size / expert_tensor_parallel_size % 128 == 0
+      precision: "fp8"
+      use_deep_gemm: true
+      gpu_memory_utilization: 0.6