Add parsed training metrics and plots

.gitattributes

@@ -34,3 +34,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text
+training_logs/20260428_191936_reward_vs_steps.png filter=lfs diff=lfs merge=lfs -text
+training_logs/20260428_191936_turn_count_distribution.png filter=lfs diff=lfs merge=lfs -text

training_logs/20260428_191936_metrics_a2_rl_stack_php_v2_400876.csv

@@ -0,0 +1,10 @@
+async/discard_rate,async/discarded_count,async/effective_batch_groups,async/effective_batch_samples,async/staleness_max,async/staleness_mean,async/staleness_min,async/staleness_ratio,generate/avg_num_tokens,generate/avg_tokens_non_zero_rewards,generate/avg_tokens_zero_rewards,generate/max_num_tokens,generate/min_num_tokens,generate/std_num_tokens,loss/avg_final_rewards,loss/avg_raw_advantages,loss/avg_raw_advantages_abs,policy/final_loss,policy/policy_entropy,policy/policy_loss,policy/policy_lr,policy/policy_update_steps,policy/ppo_clip_ratio,policy/raw_grad_norm,reward/avg_pass_at_8,reward/avg_raw_reward,system/process_rss_gb,system/process_vms_gb,system/ram_available_gb,system/ram_percent,system/ram_total_gb,system/ram_used_gb,timing/cleanup_old_checkpoints,timing/compute_advantages_and_returns,timing/convert_to_training_input,timing/fwd_logprobs_values_reward,timing/policy_train,timing/run_training,timing/save_checkpoints,timing/step,timing/sync_weights,timing/train_critic_and_policy,timing/wait_for_generation_buffer,trainer/epoch,trainer/global_step,batch_errors/total_batches,batch_errors/total_instances,batch_errors/total_successful,batch_errors/total_failed,batch_errors/total_masked,batch_errors/avg_DaytonaAuthenticationError,batch_errors/total_DaytonaAuthenticationError,batch_errors/avg_AgentEnvironmentTimeoutError,batch_errors/total_AgentEnvironmentTimeoutError,timing/save_hf_model,batch_errors/avg_DaytonaError,batch_errors/total_DaytonaError
+0.0,0,64,512,0,0.0,0,0.0,5173.5488,5581.1377,364.0,16498,1,3153.1045,0.9219,0.0054,0.0054,-0.0001,0.1612,-0.0039,0.0,1.0,0.0,0.0019,1.0,0.9219,13.9041,75.2886,363.3583,57.6,857.9687,494.6104,13.0164,0.1382,3.3894,86.136,726.5471,813.1269,45.5888,8812.9184,51.6468,726.8524,7944.7551,0,1,128,1024,896,58,126,0.015625,2.0,0.984375,126.0,,,
+0.0,0,64,512,1,1.0,1,1.0,4930.5762,5967.7683,1.0,17443,1,3568.3347,0.8262,0.0,0.0,0.0,0.1245,0.0,0.0,1.0,0.0,0.0,1.0,0.8262,15.6279,75.6176,360.3593,58.0,857.9687,497.6094,0.0131,0.0414,3.5316,74.6536,709.0029,784.0235,27.6606,3571.7276,54.3544,709.3282,2729.8153,0,2,64,512,459,21,53,,,0.828125,53.0,,,
+0.0,0,64,512,2,2.0,2,1.0,5124.7324,5696.0591,377.5273,15485,1,3123.3346,0.8926,0.0,0.0,0.0,0.1246,0.0,0.0,1.0,0.0,0.0,1.0,0.8926,15.9937,75.74,358.4,58.2,857.9687,499.5687,0.3129,0.0457,3.1477,83.4744,695.0441,778.8675,28.7517,3180.0156,51.0655,695.3471,2346.9327,0,3,64,512,472,20,40,,,0.625,40.0,,,
+0.0,0,64,512,3,2.9844,2,1.0,5176.248,5614.8284,1.0,16890,1,2922.3081,0.9219,0.0,0.0,0.0,0.115,0.0,0.0,1.0,0.0,0.0,1.0,0.9219,16.2626,75.8032,361.1512,57.9,857.9687,496.8175,0.401,0.0417,3.3777,74.1559,668.7475,743.2802,28.5269,3353.6926,50.6787,669.0823,2556.3534,0,4,64,512,473,21,39,,,0.609375,39.0,,,
+0.0,0,64,512,4,3.9375,3,1.0,5691.5176,6149.7648,284.2,26038,1,3437.1652,0.9219,-0.0008,0.0069,-0.0,0.1112,-0.0,0.0,1.0,0.0,0.0032,1.0,0.9219,21.1017,75.9924,362.8714,57.7,857.9687,495.0973,0.3297,0.06,4.5482,86.0846,752.4833,838.9829,28.3162,3308.0503,52.1042,752.838,2412.3879,0,5,64,512,505,5,7,,,0.09375,6.0,37.3861,0.015625,1.0
+0.0,0,64,512,5,4.2031,3,1.0,5741.5156,5720.9058,6532.6154,20082,1,3555.0253,0.9746,0.0,0.0,0.0,0.1169,0.0,0.0,1.0,0.0,0.0,0.9844,0.9746,21.72,75.7697,361.731,57.8,857.9687,496.2377,0.2795,0.0506,3.6853,98.4083,777.2451,876.008,32.4554,2692.5491,51.9383,777.5488,1760.8846,0,6,64,512,512,0,0,,,,,,,
+0.0,0,64,512,6,4.5781,4,1.0,10864.9395,10819.0275,18654.6667,31356,1397,9146.1479,0.9941,0.0009,0.0152,-0.0,0.1127,-0.0,0.0,1.0,0.0,0.0038,1.0,0.9941,26.2239,76.3718,428.1846,50.1,857.9687,429.7841,0.3285,0.1094,5.588,221.3386,1426.9392,1648.7818,33.3086,4928.9777,13.5134,1427.3335,3261.0614,0,7,64,512,512,0,0,,,,,,,
+0.0,0,64,512,0,0.0,0,0.0,4562.1953,4562.1953,0.0,16694,1315,2063.5003,1.0,0.0,0.0,0.0,0.1111,0.0,0.0,1.0,0.0,0.0,1.0,1.0,26.1423,76.4117,412.7117,51.9,857.9687,445.257,0.3459,0.0436,3.2475,57.5811,610.4907,668.4088,36.6974,7714.896,51.2285,610.7838,6991.9583,1,8,64,512,494,13,18,,,0.28125,18.0,,,
+0.0,0,64,512,1,1.0,1,1.0,6491.9609,6718.9148,603.1053,18269,1,3405.058,0.9629,0.0001,0.007,-0.0,0.1099,-0.0,0.0,1.0,0.0,0.0038,1.0,0.9629,26.5811,76.6696,407.6129,52.5,857.9687,450.3558,0.3266,0.0501,3.7625,93.6707,733.3147,827.3766,33.7296,2416.6198,53.5177,733.6554,1531.9602,1,9,64,512,479,20,33,,,0.515625,33.0,,,

training_logs/20260428_191936_metrics_a2_rl_stack_php_v2_400877.csv

@@ -0,0 +1,6 @@
+async/discard_rate,async/discarded_count,async/effective_batch_groups,async/effective_batch_samples,async/staleness_max,async/staleness_mean,async/staleness_min,async/staleness_ratio,generate/avg_num_tokens,generate/avg_tokens_non_zero_rewards,generate/avg_tokens_zero_rewards,generate/max_num_tokens,generate/min_num_tokens,generate/std_num_tokens,loss/avg_final_rewards,loss/avg_raw_advantages,loss/avg_raw_advantages_abs,policy/final_loss,policy/policy_entropy,policy/policy_loss,policy/policy_lr,policy/policy_update_steps,policy/ppo_clip_ratio,policy/raw_grad_norm,reward/avg_pass_at_8,reward/avg_raw_reward,system/process_rss_gb,system/process_vms_gb,system/ram_available_gb,system/ram_percent,system/ram_total_gb,system/ram_used_gb,timing/cleanup_old_checkpoints,timing/compute_advantages_and_returns,timing/convert_to_training_input,timing/fwd_logprobs_values_reward,timing/policy_train,timing/run_training,timing/save_checkpoints,timing/save_hf_model,timing/step,timing/sync_weights,timing/train_critic_and_policy,timing/wait_for_generation_buffer,trainer/epoch,trainer/global_step,batch_errors/total_batches,batch_errors/total_instances,batch_errors/total_successful,batch_errors/total_failed,batch_errors/total_masked,batch_errors/avg_AgentEnvironmentTimeoutError,batch_errors/total_AgentEnvironmentTimeoutError,batch_errors/avg_DaytonaError,batch_errors/total_DaytonaError
+0.0,0,64,512,2,0.4375,0,0.2188,4768.9824,4805.7866,1665.1667,18141,1,2456.2799,0.9883,-0.0007,0.0075,0.0,0.1119,0.0,0.0,1.0,0.0,0.0033,1.0,0.9883,15.2558,75.1672,370.8639,56.8,857.9687,487.1048,8.2923,0.0521,3.272,74.7888,683.9612,759.0926,40.7566,42.2577,7720.7893,64.0786,684.2511,6894.3459,1,10,64,512,470,22,42,0.65625,42.0,,
+0.0,0,64,512,1,1.0,1,1.0,6043.1543,6529.5032,1231.4043,14905,1,3379.2364,0.9082,0.001,0.0085,0.0,0.1094,0.0,0.0,1.0,0.0,0.0039,1.0,0.9082,15.4861,75.2612,369.6674,56.9,857.9687,488.3013,0.2726,0.0386,3.2801,86.1438,735.4492,822.0201,33.6901,,1944.634,52.3307,735.837,1066.9959,1,11,64,512,477,20,35,0.546875,35.0,,
+0.0,0,64,512,2,2.0,2,1.0,5200.8203,5582.3585,1.0,18761,1,3026.5222,0.9316,0.0,0.0,0.0,0.1061,0.0,0.0,1.0,0.0,0.0,1.0,0.9316,15.956,75.5988,372.0403,56.6,857.9687,485.9283,0.315,0.0443,3.8178,83.3515,683.5771,767.262,34.4038,,3377.1913,53.2947,683.8658,2552.7951,1,12,64,512,500,6,12,0.1875,12.0,,
+0.0,0,64,512,3,3.0,3,1.0,5112.4043,5205.6962,2021.3333,20402,1,2908.5829,0.9707,-0.0012,0.0154,-0.0,0.1116,-0.0,0.0,1.0,0.0,0.0057,1.0,0.9707,18.1414,75.7131,370.1971,56.9,857.9687,487.7716,0.3715,0.0475,3.4987,78.1533,686.8594,765.3441,34.4024,,2811.1464,52.3944,687.143,1989.8831,1,13,64,512,506,5,6,0.078125,5.0,0.015625,1.0
+0.0,0,64,512,4,4.0,4,1.0,9612.0879,9534.167,12193.8667,30017,1,7895.0271,0.9707,-0.0011,0.0202,-0.0,0.1125,-0.0,0.0,1.0,0.0,0.0025,0.9844,0.9707,25.5219,76.0509,411.2549,52.1,857.9687,446.7137,0.2489,0.0637,5.1805,202.8086,1219.5955,1422.8569,33.357,,3814.0265,48.8945,1219.9843,2337.0674,1,14,12,96,96,0,0,,,,

training_logs/20260428_191936_metrics_report.md

@@ -0,0 +1,263 @@
+# SkyRL Training Metrics Analysis
+
+Generated from 2 log files
+
+## Overview
+
+| Log File | Total Steps | Metric Blocks | Final Reward (mean) | Final Reward (max) | Total Time (s) |
+|----------|-------------|---------------|---------------------|-------------------|----------------|
+| a2_rl_stack_php_v2_400876 | 9 | 9 | 0.9351 | 1.0000 | 39979.4 |
+| a2_rl_stack_php_v2_400877 | 14 | 5 | 0.9539 | 0.9883 | 19667.8 |
+
+## Async Metrics
+
+|                               |       Mean |      Std |   Min |      Max |   Count |
+|:------------------------------|-----------:|---------:|------:|---------:|--------:|
+| async/discard_rate            |   0        | 0        |     0 |   0      |      14 |
+| async/discarded_count         |   0        | 0        |     0 |   0      |      14 |
+| async/effective_batch_groups  |  64        | 0        |    64 |  64      |      14 |
+| async/effective_batch_samples | 512        | 0        |   512 | 512      |      14 |
+| async/staleness_max           |   2.42857  | 1.82775  |     0 |   6      |      14 |
+| async/staleness_mean          |   2.1529   | 1.62522  |     0 |   4.5781 |      14 |
+| async/staleness_min           |   1.85714  | 1.40642  |     0 |   4      |      14 |
+| async/staleness_ratio         |   0.801343 | 0.397857 |     0 |   1      |      14 |
+
+## Generate Metrics
+
+|                                      |      Mean |      Std |     Min |      Max |   Count |
+|:-------------------------------------|----------:|---------:|--------:|---------:|--------:|
+| generate/avg_num_tokens              |  6035.33  | 1868.44  |  4562.2 | 10864.9  |      14 |
+| generate/avg_tokens_non_zero_rewards |  6320.58  | 1751.66  |  4562.2 | 10819    |      14 |
+| generate/avg_tokens_zero_rewards     |  3137.92  | 5620.05  |     0   | 18654.7  |      14 |
+| generate/max_num_tokens              | 20070.1   | 5259.36  | 14905   | 31356    |      14 |
+| generate/min_num_tokens              |   194.571 |  492.313 |     1   |  1397    |      14 |
+| generate/std_num_tokens              |  3859.97  | 2033.17  |  2063.5 |  9146.15 |      14 |
+
+## Loss Metrics
+
+|                             |        Mean |        Std |     Min |    Max |   Count |
+|:----------------------------|------------:|-----------:|--------:|-------:|--------:|
+| loss/avg_final_rewards      | 0.941829    | 0.0477826  |  0.8262 | 1      |      14 |
+| loss/avg_raw_advantages     | 0.000257143 | 0.00161327 | -0.0012 | 0.0054 |      14 |
+| loss/avg_raw_advantages_abs | 0.00615     | 0.00679929 |  0      | 0.0202 |      14 |
+
+## Policy Metrics
+
+|                            |         Mean |         Std |     Min |    Max |   Count |
+|:---------------------------|-------------:|------------:|--------:|-------:|--------:|
+| policy/final_loss          | -7.14286e-06 | 2.67261e-05 | -0.0001 | 0      |      14 |
+| policy/policy_entropy      |  0.117043    | 0.0137588   |  0.1061 | 0.1612 |      14 |
+| policy/policy_loss         | -0.000278571 | 0.00104232  | -0.0039 | 0      |      14 |
+| policy/policy_lr           |  0           | 0           |  0      | 0      |      14 |
+| policy/policy_update_steps |  1           | 0           |  1      | 1      |      14 |
+| policy/ppo_clip_ratio      |  0           | 0           |  0      | 0      |      14 |
+| policy/raw_grad_norm       |  0.00200714  | 0.00198396  |  0      | 0.0057 |      14 |
+
+## Reward Metrics
+
+|                       |     Mean |        Std |    Min |   Max |   Count |
+|:----------------------|---------:|-----------:|-------:|------:|--------:|
+| reward/avg_pass_at_8  | 0.997771 | 0.00566493 | 0.9844 |     1 |      14 |
+| reward/avg_raw_reward | 0.941829 | 0.0477826  | 0.8262 |     1 |      14 |
+
+## System Metrics
+
+|                         |     Mean |          Std |      Min |      Max |   Count |
+|:------------------------|---------:|-------------:|---------:|---------:|--------:|
+| system/process_rss_gb   |  19.5656 |  4.8101      |  13.9041 |  26.5811 |      14 |
+| system/process_vms_gb   |  75.8183 |  0.446416    |  75.1672 |  76.6696 |      14 |
+| system/ram_available_gb | 379.315  | 24.1565      | 358.4    | 428.185  |      14 |
+| system/ram_percent      |  55.7857 |  2.8038      |  50.1    |  58.2    |      14 |
+| system/ram_total_gb     | 857.969  |  2.35957e-13 | 857.969  | 857.969  |      14 |
+| system/ram_used_gb      | 478.654  | 24.1565      | 429.784  | 499.569  |      14 |
+
+## Timing Metrics
+
+|                                       |         Mean |          Std |       Min |       Max |   Count |
+|:--------------------------------------|-------------:|-------------:|----------:|----------:|--------:|
+| timing/cleanup_old_checkpoints        |    1.77528   |    3.87516   |    0.0131 |   13.0164 |      14 |
+| timing/compute_advantages_and_returns |    0.0590643 |    0.0288508 |    0.0386 |    0.1382 |      14 |
+| timing/convert_to_training_input      |    3.80907   |    0.757823  |    3.1477 |    5.588  |      14 |
+| timing/fwd_logprobs_values_reward     |  100.054     |   48.5845    |   57.5811 |  221.339  |      14 |
+| timing/policy_train                   |  793.518     |  231.644     |  610.491  | 1426.94   |      14 |
+| timing/run_training                   |  893.959     |  279.927     |  668.409  | 1648.78   |      14 |
+| timing/save_checkpoints               |   33.6889    |    4.95858   |   27.6606 |   45.5888 |      14 |
+| timing/step                           | 4260.52      | 2197.04      | 1944.63   | 8812.92   |      14 |
+| timing/sync_weights                   |   50.0743    |   11.087     |   13.5134 |   64.0786 |      14 |
+| timing/train_critic_and_policy        |  793.846     |  231.673     |  610.784  | 1427.33   |      14 |
+| timing/wait_for_generation_buffer     | 3312.66      | 2225.38      | 1067      | 7944.76   |      14 |
+| timing/save_hf_model                  |   39.8219    |    3.44474   |   37.3861 |   42.2577 |       2 |
+
+## Trainer Metrics
+
+|                     |   Mean |      Std |   Min |   Max |   Count |
+|:--------------------|-------:|---------:|------:|------:|--------:|
+| trainer/epoch       |    0.5 | 0.518875 |     0 |     1 |      14 |
+| trainer/global_step |    7.5 | 4.1833   |     1 |    14 |      14 |
+
+## Batch_Errors Metrics
+
+|                                                 |       Mean |        Std |       Min |         Max |   Count |
+|:------------------------------------------------|-----------:|-----------:|----------:|------------:|--------:|
+| batch_errors/total_batches                      |  64.8571   |  22.8536   | 12        |  128        |      14 |
+| batch_errors/total_instances                    | 518.857    | 182.829    | 96        | 1024        |      14 |
+| batch_errors/total_successful                   | 489.357    | 157.845    | 96        |  896        |      14 |
+| batch_errors/total_failed                       |  15.0714   |  15.2036   |  0        |   58        |      14 |
+| batch_errors/total_masked                       |  29.3571   |  33.2487   |  0        |  126        |      14 |
+| batch_errors/avg_DaytonaAuthenticationError     |   0.015625 | nan        |  0.015625 |    0.015625 |       1 |
+| batch_errors/total_DaytonaAuthenticationError   |   2        | nan        |  2        |    2        |       1 |
+| batch_errors/avg_AgentEnvironmentTimeoutError   |   0.491477 |   0.297398 |  0.078125 |    0.984375 |      11 |
+| batch_errors/total_AgentEnvironmentTimeoutError |  37.1818   |  33.4748   |  5        |  126        |      11 |
+| batch_errors/avg_DaytonaError                   |   0.015625 |   0        |  0.015625 |    0.015625 |       2 |
+| batch_errors/total_DaytonaError                 |   1        |   0        |  1        |    1        |       2 |
+
+## Training Progression by Log
+
+### a2_rl_stack_php_v2_400876
+
+| Step | Reward | Pass@8 | KL | Loss | Step Time (s) | Gen Wait (s) |
+|------|--------|--------|-----|------|---------------|-------------|
+| 1 | 0.9219 | 1.0000 | 0.000000 | -0.0001 | 8812.9 | 7944.8 |
+| 2 | 0.8262 | 1.0000 | 0.000000 | 0.0000 | 3571.7 | 2729.8 |
+| 3 | 0.8926 | 1.0000 | 0.000000 | 0.0000 | 3180.0 | 2346.9 |
+| 4 | 0.9219 | 1.0000 | 0.000000 | 0.0000 | 3353.7 | 2556.4 |
+| 5 | 0.9219 | 1.0000 | 0.000000 | -0.0000 | 3308.1 | 2412.4 |
+| 6 | 0.9746 | 0.9844 | 0.000000 | 0.0000 | 2692.5 | 1760.9 |
+| 7 | 0.9941 | 1.0000 | 0.000000 | -0.0000 | 4929.0 | 3261.1 |
+| 8 | 1.0000 | 1.0000 | 0.000000 | 0.0000 | 7714.9 | 6992.0 |
+| 9 | 0.9629 | 1.0000 | 0.000000 | -0.0000 | 2416.6 | 1532.0 |
+
+### a2_rl_stack_php_v2_400877
+
+| Step | Reward | Pass@8 | KL | Loss | Step Time (s) | Gen Wait (s) |
+|------|--------|--------|-----|------|---------------|-------------|
+| 10 | 0.9883 | 1.0000 | 0.000000 | 0.0000 | 7720.8 | 6894.3 |
+| 11 | 0.9082 | 1.0000 | 0.000000 | 0.0000 | 1944.6 | 1067.0 |
+| 12 | 0.9316 | 1.0000 | 0.000000 | 0.0000 | 3377.2 | 2552.8 |
+| 13 | 0.9707 | 1.0000 | 0.000000 | -0.0000 | 2811.1 | 1989.9 |
+| 14 | 0.9707 | 0.9844 | 0.000000 | -0.0000 | 3814.0 | 2337.1 |
+
+## Timing Analysis
+
+### Average Time Breakdown (% of step time)
+
+| Component | Avg % of Step Time |
+|-----------|-------------------|
+| wait_for_generation_buffer | 73.4% |
+| run_training | 25.1% |
+| train_critic_and_policy | 22.3% |
+| policy_train | 22.3% |
+| fwd_logprobs_values_reward | 2.8% |
+| sync_weights | 1.4% |
+| save_checkpoints | 0.9% |
+| save_hf_model | 0.8% |
+| convert_to_training_input | 0.1% |
+| cleanup_old_checkpoints | 0.0% |
+| compute_advantages_and_returns | 0.0% |
+
+## Cross-Log Comparison
+
+| Log | Avg Reward | Pass@8 | Step Time (s) | Gen Wait Time (s) | Avg Tokens | Staleness |
+|-----|------|------|------|------|------|------|
+| a2_rl_stack_php_v2_400876 | 0.9351 | 0.9983 | 4442.1608 | 3504.0121 | 5973.0260 | 2.1892 |
+| a2_rl_stack_php_v2_400877 | 0.9539 | 0.9969 | 3933.5575 | 2968.2175 | 6147.4898 | 2.0875 |
+
+## vLLM Inference Engine Analysis
+
+Metrics from vLLM stat loggers (V1LoggingStatLoggerFixed).
+
+> **Note**: Ray deduplicates similar log messages with `[repeated Nx across cluster]`,
+> so we typically capture stats from one engine per timestamp. The stats shown are
+> **per-engine** values. Multiply by num_inference_engines for cluster-wide estimates.
+
+### Summary by Log (Per-Engine Stats)
+
+| Log | Avg Running/Engine | Avg Waiting/Engine | Avg Gen Throughput/Engine | Avg KV Cache % | Avg Prefix Hit % |
+|-----|-------------------|-------------------|--------------------------|----------------|------------------|
+| a2_rl_stack_php_v2_400876 | 4.5 | 0.0 | 92.5 tok/s | 8.2% | 85.3% |
+| a2_rl_stack_php_v2_400877 | 4.3 | 0.0 | 87.0 tok/s | 8.6% | 86.3% |
+
+### Utilization Analysis (Per-Engine)
+
+Key indicators of inference engine utilization:
+
+- **Running requests/engine**: Concurrent requests being processed by each engine
+- **Waiting requests**: Requests queued (0 = engine not saturated, has spare capacity)
+- **Generation throughput**: Decode tokens/sec per engine
+  - 8B model on H100 can do **1000+ tok/s** when saturated
+  - If seeing <300 tok/s with 0 waiting, engine is **starved for requests**
+
+#### a2_rl_stack_php_v2_400876
+
+- **Running requests/engine**: avg=4.5, max=16
+- **Waiting requests**: avg=0.0, max=0
+- **Generation throughput/engine**: avg=92.5 tok/s, max=348.4 tok/s
+- **KV cache usage**: avg=8.2%
+- **Prefix cache hit rate**: avg=85.3%
+- ⚠️ **Underutilized**: Engines starved for requests (0 waiting, avg 4.5 running)
+  - Bottleneck is likely upstream (environment execution, not inference)
+
+#### a2_rl_stack_php_v2_400877
+
+- **Running requests/engine**: avg=4.3, max=15
+- **Waiting requests**: avg=0.0, max=0
+- **Generation throughput/engine**: avg=87.0 tok/s, max=307.0 tok/s
+- **KV cache usage**: avg=8.6%
+- **Prefix cache hit rate**: avg=86.3%
+- ⚠️ **Underutilized**: Engines starved for requests (0 waiting, avg 4.3 running)
+  - Bottleneck is likely upstream (environment execution, not inference)
+
+## Trial-Level Analysis (from result.json)
+
+Total trials parsed: 8273
+
+### Turn Count Statistics
+
+| Metric | Value |
+|--------|-------|
+| Mean | 3.8 |
+| Median | 3.0 |
+| Std | 3.1 |
+| Min | 1 |
+| Max | 62 |
+| Count | 8273 |
+
+### Exception Distribution
+
+| Exception Type | Count | % |
+|---------------|-------|---|
+| No exception | 6920 | 83.6% |
+| AgentTimeoutError | 733 | 8.9% |
+| AgentEnvironmentTimeoutError | 410 | 5.0% |
+| ContextLengthExceededError | 147 | 1.8% |
+| VerifierRuntimeError | 58 | 0.7% |
+| DaytonaError | 2 | 0.0% |
+| DaytonaAuthenticationError | 2 | 0.0% |
+| InternalServerError | 1 | 0.0% |
+
+### Turn Count by Exception Type
+
+| Exception Type | Mean Turns | Median Turns | Count |
+|---------------|-----------|-------------|-------|
+| InternalServerError | 22.0 | 22.0 | 1 |
+| ContextLengthExceededError | 20.9 | 20.0 | 147 |
+| AgentTimeoutError | 7.0 | 6.0 | 733 |
+| AgentEnvironmentTimeoutError | 4.9 | 5.0 | 410 |
+| VerifierRuntimeError | 4.8 | 5.0 | 58 |
+| No exception | 3.0 | 3.0 | 6920 |
+| DaytonaError | 2.0 | 2.0 | 2 |
+| DaytonaAuthenticationError | 1.0 | 1.0 | 2 |
+
+### Turn Count by Outcome
+
+| Outcome | Mean Turns | Median Turns | Count |
+|---------|-----------|-------------|-------|
+| Success | 3.8 | 3.0 | 8162 |
+| Failure | 7.9 | 7.0 | 47 |
+
+### Reward Summary
+
+- Mean reward: 0.9943
+- Success rate: 99.4%
+- Trials with reward data: 8209
+

training_logs/20260428_191936_metrics_table.csv

@@ -0,0 +1,15 @@
+log_file,async/discard_rate,async/discarded_count,async/effective_batch_groups,async/effective_batch_samples,async/staleness_max,async/staleness_mean,async/staleness_min,async/staleness_ratio,generate/avg_num_tokens,generate/avg_tokens_non_zero_rewards,generate/avg_tokens_zero_rewards,generate/max_num_tokens,generate/min_num_tokens,generate/std_num_tokens,loss/avg_final_rewards,loss/avg_raw_advantages,loss/avg_raw_advantages_abs,policy/final_loss,policy/policy_entropy,policy/policy_loss,policy/policy_lr,policy/policy_update_steps,policy/ppo_clip_ratio,policy/raw_grad_norm,reward/avg_pass_at_8,reward/avg_raw_reward,system/process_rss_gb,system/process_vms_gb,system/ram_available_gb,system/ram_percent,system/ram_total_gb,system/ram_used_gb,timing/cleanup_old_checkpoints,timing/compute_advantages_and_returns,timing/convert_to_training_input,timing/fwd_logprobs_values_reward,timing/policy_train,timing/run_training,timing/save_checkpoints,timing/step,timing/sync_weights,timing/train_critic_and_policy,timing/wait_for_generation_buffer,trainer/epoch,trainer/global_step,batch_errors/total_batches,batch_errors/total_instances,batch_errors/total_successful,batch_errors/total_failed,batch_errors/total_masked,batch_errors/avg_DaytonaAuthenticationError,batch_errors/total_DaytonaAuthenticationError,batch_errors/avg_AgentEnvironmentTimeoutError,batch_errors/total_AgentEnvironmentTimeoutError,timing/save_hf_model,batch_errors/avg_DaytonaError,batch_errors/total_DaytonaError,global_step
+a2_rl_stack_php_v2_400876,0.0,0,64,512,0,0.0,0,0.0,5173.5488,5581.1377,364.0,16498,1,3153.1045,0.9219,0.0054,0.0054,-0.0001,0.1612,-0.0039,0.0,1.0,0.0,0.0019,1.0,0.9219,13.9041,75.2886,363.3583,57.6,857.9687,494.6104,13.0164,0.1382,3.3894,86.136,726.5471,813.1269,45.5888,8812.9184,51.6468,726.8524,7944.7551,0,1,128,1024,896,58,126,0.015625,2.0,0.984375,126.0,,,,1
+a2_rl_stack_php_v2_400876,0.0,0,64,512,1,1.0,1,1.0,4930.5762,5967.7683,1.0,17443,1,3568.3347,0.8262,0.0,0.0,0.0,0.1245,0.0,0.0,1.0,0.0,0.0,1.0,0.8262,15.6279,75.6176,360.3593,58.0,857.9687,497.6094,0.0131,0.0414,3.5316,74.6536,709.0029,784.0235,27.6606,3571.7276,54.3544,709.3282,2729.8153,0,2,64,512,459,21,53,,,0.828125,53.0,,,,2
+a2_rl_stack_php_v2_400876,0.0,0,64,512,2,2.0,2,1.0,5124.7324,5696.0591,377.5273,15485,1,3123.3346,0.8926,0.0,0.0,0.0,0.1246,0.0,0.0,1.0,0.0,0.0,1.0,0.8926,15.9937,75.74,358.4,58.2,857.9687,499.5687,0.3129,0.0457,3.1477,83.4744,695.0441,778.8675,28.7517,3180.0156,51.0655,695.3471,2346.9327,0,3,64,512,472,20,40,,,0.625,40.0,,,,3
+a2_rl_stack_php_v2_400876,0.0,0,64,512,3,2.9844,2,1.0,5176.248,5614.8284,1.0,16890,1,2922.3081,0.9219,0.0,0.0,0.0,0.115,0.0,0.0,1.0,0.0,0.0,1.0,0.9219,16.2626,75.8032,361.1512,57.9,857.9687,496.8175,0.401,0.0417,3.3777,74.1559,668.7475,743.2802,28.5269,3353.6926,50.6787,669.0823,2556.3534,0,4,64,512,473,21,39,,,0.609375,39.0,,,,4
+a2_rl_stack_php_v2_400876,0.0,0,64,512,4,3.9375,3,1.0,5691.5176,6149.7648,284.2,26038,1,3437.1652,0.9219,-0.0008,0.0069,-0.0,0.1112,-0.0,0.0,1.0,0.0,0.0032,1.0,0.9219,21.1017,75.9924,362.8714,57.7,857.9687,495.0973,0.3297,0.06,4.5482,86.0846,752.4833,838.9829,28.3162,3308.0503,52.1042,752.838,2412.3879,0,5,64,512,505,5,7,,,0.09375,6.0,37.3861,0.015625,1.0,5
+a2_rl_stack_php_v2_400876,0.0,0,64,512,5,4.2031,3,1.0,5741.5156,5720.9058,6532.6154,20082,1,3555.0253,0.9746,0.0,0.0,0.0,0.1169,0.0,0.0,1.0,0.0,0.0,0.9844,0.9746,21.72,75.7697,361.731,57.8,857.9687,496.2377,0.2795,0.0506,3.6853,98.4083,777.2451,876.008,32.4554,2692.5491,51.9383,777.5488,1760.8846,0,6,64,512,512,0,0,,,,,,,,6
+a2_rl_stack_php_v2_400876,0.0,0,64,512,6,4.5781,4,1.0,10864.9395,10819.0275,18654.6667,31356,1397,9146.1479,0.9941,0.0009,0.0152,-0.0,0.1127,-0.0,0.0,1.0,0.0,0.0038,1.0,0.9941,26.2239,76.3718,428.1846,50.1,857.9687,429.7841,0.3285,0.1094,5.588,221.3386,1426.9392,1648.7818,33.3086,4928.9777,13.5134,1427.3335,3261.0614,0,7,64,512,512,0,0,,,,,,,,7
+a2_rl_stack_php_v2_400876,0.0,0,64,512,0,0.0,0,0.0,4562.1953,4562.1953,0.0,16694,1315,2063.5003,1.0,0.0,0.0,0.0,0.1111,0.0,0.0,1.0,0.0,0.0,1.0,1.0,26.1423,76.4117,412.7117,51.9,857.9687,445.257,0.3459,0.0436,3.2475,57.5811,610.4907,668.4088,36.6974,7714.896,51.2285,610.7838,6991.9583,1,8,64,512,494,13,18,,,0.28125,18.0,,,,8
+a2_rl_stack_php_v2_400876,0.0,0,64,512,1,1.0,1,1.0,6491.9609,6718.9148,603.1053,18269,1,3405.058,0.9629,0.0001,0.007,-0.0,0.1099,-0.0,0.0,1.0,0.0,0.0038,1.0,0.9629,26.5811,76.6696,407.6129,52.5,857.9687,450.3558,0.3266,0.0501,3.7625,93.6707,733.3147,827.3766,33.7296,2416.6198,53.5177,733.6554,1531.9602,1,9,64,512,479,20,33,,,0.515625,33.0,,,,9
+a2_rl_stack_php_v2_400877,0.0,0,64,512,2,0.4375,0,0.2188,4768.9824,4805.7866,1665.1667,18141,1,2456.2799,0.9883,-0.0007,0.0075,0.0,0.1119,0.0,0.0,1.0,0.0,0.0033,1.0,0.9883,15.2558,75.1672,370.8639,56.8,857.9687,487.1048,8.2923,0.0521,3.272,74.7888,683.9612,759.0926,40.7566,7720.7893,64.0786,684.2511,6894.3459,1,10,64,512,470,22,42,,,0.65625,42.0,42.2577,,,10
+a2_rl_stack_php_v2_400877,0.0,0,64,512,1,1.0,1,1.0,6043.1543,6529.5032,1231.4043,14905,1,3379.2364,0.9082,0.001,0.0085,0.0,0.1094,0.0,0.0,1.0,0.0,0.0039,1.0,0.9082,15.4861,75.2612,369.6674,56.9,857.9687,488.3013,0.2726,0.0386,3.2801,86.1438,735.4492,822.0201,33.6901,1944.634,52.3307,735.837,1066.9959,1,11,64,512,477,20,35,,,0.546875,35.0,,,,11
+a2_rl_stack_php_v2_400877,0.0,0,64,512,2,2.0,2,1.0,5200.8203,5582.3585,1.0,18761,1,3026.5222,0.9316,0.0,0.0,0.0,0.1061,0.0,0.0,1.0,0.0,0.0,1.0,0.9316,15.956,75.5988,372.0403,56.6,857.9687,485.9283,0.315,0.0443,3.8178,83.3515,683.5771,767.262,34.4038,3377.1913,53.2947,683.8658,2552.7951,1,12,64,512,500,6,12,,,0.1875,12.0,,,,12
+a2_rl_stack_php_v2_400877,0.0,0,64,512,3,3.0,3,1.0,5112.4043,5205.6962,2021.3333,20402,1,2908.5829,0.9707,-0.0012,0.0154,-0.0,0.1116,-0.0,0.0,1.0,0.0,0.0057,1.0,0.9707,18.1414,75.7131,370.1971,56.9,857.9687,487.7716,0.3715,0.0475,3.4987,78.1533,686.8594,765.3441,34.4024,2811.1464,52.3944,687.143,1989.8831,1,13,64,512,506,5,6,,,0.078125,5.0,,0.015625,1.0,13
+a2_rl_stack_php_v2_400877,0.0,0,64,512,4,4.0,4,1.0,9612.0879,9534.167,12193.8667,30017,1,7895.0271,0.9707,-0.0011,0.0202,-0.0,0.1125,-0.0,0.0,1.0,0.0,0.0025,0.9844,0.9707,25.5219,76.0509,411.2549,52.1,857.9687,446.7137,0.2489,0.0637,5.1805,202.8086,1219.5955,1422.8569,33.357,3814.0265,48.8945,1219.9843,2337.0674,1,14,12,96,96,0,0,,,,,,,,14