metrics check

bg51717 · bg51717 · commit 50e878e4b81b · 2026-05-21T03:28:39.000-07:00
Signed-off-by: bg51717 &lt;biguo@nvidia.com&gt;
diff --git a/docs/index.md b/docs/index.md
@@ -62,7 +62,7 @@ Learn about DTensor and Megatron Core training backends, their capabilities, and
 :link: about/algorithms/index
 :link-type: doc
 
-Discover supported algorithms including GRPO, SFT, DPO, RM, and on-policy distillation with detailed guides and examples.
+Discover supported algorithms including GRPO, PPO, SFT, DPO, RM, and on-policy distillation with detailed guides and examples.
 :::
 
 :::{grid-item-card} {octicon}`graph` Evaluation
@@ -239,6 +239,7 @@ guides/dpo.md
 guides/dapo.md
 guides/prorlv2.md
 guides/grpo.md
+guides/ppo.md
 guides/grpo-deepscaler.md
 guides/grpo-sliding-puzzle.md
 guides/grpo-audio.md
diff --git a/tests/functional/L1_Functional_Tests_GPU.sh b/tests/functional/L1_Functional_Tests_GPU.sh
@@ -68,6 +68,7 @@ run_test      uv run --no-sync bash ./tests/functional/grpo_non_colocated.sh
 run_test      uv run --no-sync bash ./tests/functional/grpo_rm_env.sh
 run_test      uv run --no-sync bash ./tests/functional/grpo_sglang.sh
 run_test fast uv run --no-sync bash ./tests/functional/grpo_topp_topk.sh
+run_test      uv run --no-sync bash ./tests/functional/ppo_megatron.sh
 run_test      uv run --no-sync bash ./tests/functional/prorlv2.sh
 run_test      uv run --no-sync bash ./tests/functional/qa_distillation_megatron.sh
 run_test      uv run --no-sync bash ./tests/functional/rm.sh
diff --git a/tests/test_suites/llm/ppo-dsr1-7b-math-8n8g-megatron.sh b/tests/test_suites/llm/ppo-dsr1-7b-math-8n8g-megatron.sh
@@ -34,9 +34,8 @@ uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
 # Only run metrics if the target step is reached
 if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
     uv run tests/check_metrics.py $JSON_METRICS \
-        'median(data["train/token_mult_prob_error"]) < 1.1' \
-        'data["train/token_mult_prob_error"]["20"] < 1.1' \
-        'data["train/reward"]["20"] > 0.0'
+        'data["train/reward"]["40"] > 0.3' \
+        'data["validation/accuracy"]["40"] > 0.42'
 
     # Clean up checkpoint directory after successful run to save space.
     rm -rf "$CKPT_DIR"
diff --git a/tests/test_suites/llm/ppo-qwen2.5-1.5b-gsm8k-1n8g-megatron.sh b/tests/test_suites/llm/ppo-qwen2.5-1.5b-gsm8k-1n8g-megatron.sh
@@ -34,9 +34,8 @@ uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
 # Only run metrics if the target step is reached
 if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
     uv run tests/check_metrics.py $JSON_METRICS \
-        'median(data["train/token_mult_prob_error"]) < 1.1' \
-        'data["train/token_mult_prob_error"]["100"] < 1.1' \
-        'data["train/reward"]["100"] > 0.5'
+        'data["train/reward"]["100"] > 0.85' \
+        'data["validation/accuracy"]["100"] > 0.7'
 
     # Clean up checkpoint directory after successful run to save space.
     rm -rf "$CKPT_DIR"
diff --git a/tests/test_suites/nightly.txt b/tests/test_suites/nightly.txt
@@ -173,8 +173,8 @@ tests/test_suites/llm/distillation-qwen3-1.7b-1n8g-megatron-qa-nvfp4.sh
 # PPO #
 #######
 
-# Megatron PPO with GAE and value model (DeepSeek-7B, GSM8K)
-tests/test_suites/llm/ppo-deepseek7b-gsm8k-1n8g-megatron.sh
+# Megatron PPO with GAE and value model (Qwen2.5-1.5B, GSM8K)
+tests/test_suites/llm/ppo-qwen2.5-1.5b-gsm8k-1n8g-megatron.sh
 
 # Megatron PPO with KL penalty + importance sampling (DeepSeek-R1-7B, DAPOMath)
 tests/test_suites/llm/ppo-dsr1-7b-math-8n8g-megatron.sh
diff --git a/tests/unit/reference_configs/ppo_math_1B_megatron.yaml b/tests/unit/reference_configs/ppo_math_1B_megatron.yaml
