fix: grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts runs 40 steps (#1231)

Signed-off-by: Terry Kong <terryk@nvidia.com>

Author: terrykong

…instruct-1n8g-megatron-fp8-rollouts.yaml …truct-1n8g-megatron-fp8-rollouts.v2.yamlexamples/configs/recipes/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.yaml renamed to examples/configs/recipes/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.v2.yaml examples/configs/recipes/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.yaml renamed to examples/configs/recipes/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.v2.yaml

@@ -42,11 +42,11 @@ policy:
 data:
   max_input_seq_length: 4096
 logger:
-  log_dir: logs/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts
+  log_dir: logs/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.v2
   wandb_enabled: true
   tensorboard_enabled: true
   wandb:
     project: nemo-rl
-    name: grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts
+    name: grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.v2
 cluster:
   gpus_per_node: 8

…b-instruct-1n8g-megatron-fp8-rollouts.sh …nstruct-1n8g-megatron-fp8-rollouts.v2.shtests/test_suites/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.sh renamed to tests/test_suites/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.v2.sh tests/test_suites/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.sh renamed to tests/test_suites/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.v2.sh

@@ -4,8 +4,8 @@ source $SCRIPT_DIR/common.env
 
 # ===== BEGIN CONFIG =====
 NUM_NODES=1
-STEPS_PER_RUN=100
-MAX_STEPS=100
+STEPS_PER_RUN=40
+MAX_STEPS=40
 NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
 NUM_MINUTES=120
 # ===== END CONFIG =====

tests/test_suites/nightly.txt

@@ -35,7 +35,7 @@ tests/test_suites/llm/grpo-gspo-deepscaler-1.5b-8K.sh
 tests/test_suites/llm/grpo-math-qwen3-30ba3b-megatron-tp4-32k.sh
 
 # FP8
-tests/test_suites/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.sh
+tests/test_suites/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-rollouts.v2.sh
 tests/test_suites/llm/grpo-llama3.1-8b-instruct-1n8g-megatron-fp8-e2e.sh
 
 # Non-colocated