Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76780.75659017764,
-      "required_pages_per_second_upper_estimate": 0.005311870896808872
+      "required_token_positions_per_second": 77146.356571443,
+      "required_pages_per_second_upper_estimate": 0.0053371639518216395
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9597.594573772205,
-      "required_pages_per_second_upper_estimate": 0.00401831242747314
+      "required_token_positions_per_second": 9643.294571430375,
+      "required_pages_per_second_upper_estimate": 0.004037446062168145
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19195.18914754441,
-      "required_pages_per_second_upper_estimate": 0.003997851688254453
+      "required_token_positions_per_second": 19286.58914286075,
+      "required_pages_per_second_upper_estimate": 0.004016887896898878
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28792.783721316613,
-      "required_pages_per_second_upper_estimate": 0.0017260834200577887
+      "required_token_positions_per_second": 28929.883714291125,
+      "required_pages_per_second_upper_estimate": 0.0017343023552970425
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28792.783721316613,
-      "required_pages_per_second_upper_estimate": 0.002238512288261146
+      "required_token_positions_per_second": 28929.883714291125,
+      "required_pages_per_second_upper_estimate": 0.002249171209675776
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28792.783721316613,
-      "required_pages_per_second_upper_estimate": 0.36932291429453973
+      "required_token_positions_per_second": 28929.883714291125,
+      "required_pages_per_second_upper_estimate": 0.3710814857979134
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1442483061252915,
-    "producer_p95_seconds": 0.33357249200344086,
-    "gpu_update_seconds": 0.9602406029095034,
-    "producer_batches_per_second": 6.932490417817578,
-    "gpu_batches_per_second": 1.041405661216602,
-    "producer_headroom": 6.656858778469507,
+    "producer_batch_seconds": 0.09749988875779536,
+    "producer_p95_seconds": 0.3350607320026029,
+    "gpu_update_seconds": 0.9556899804039695,
+    "producer_batches_per_second": 10.256421958430671,
+    "gpu_batches_per_second": 1.0463644283236084,
+    "producer_headroom": 9.80195970046745,
     "sustainable": true,
-    "buffer_seconds": 60.49515798329872,
+    "buffer_seconds": 60.20846876545008,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 247072.63852710513,
-    "upload_bytes_per_second": 5145160.357073876,
+    "generation_bytes_per_second": 264801.0249412577,
+    "upload_bytes_per_second": 5096453.290684326,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.01603667125573,
+    "estimated_full_shards_per_hour": 56.820135700018874,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.046063713998592,
-      "bytes": 5944086,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.323775843000476,
       "bytes": 7235009,
@@ -562,6 +556,12 @@
       "bytes": 7586487,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.84151389899489,
+      "bytes": 763221,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -398,3 +398,4 @@
 {"step": 924800, "loss": 0.5135881900787354, "lr": 3e-05, "grad_norm": 0.34735354490061177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135364418314, "tokens_per_second": 140280.27799993314, "data_wait_s": 1.615499786566943e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 924900, "loss": 0.5153971910476685, "lr": 3e-05, "grad_norm": 0.3553751987668995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135379093364, "tokens_per_second": 152545.82118496113, "data_wait_s": 1.914300082717091e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 925000, "loss": 0.5517297387123108, "lr": 3e-05, "grad_norm": 0.34564684448204536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135393509500, "tokens_per_second": 150325.96381885314, "data_wait_s": 9.957002475857735e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 925100, "loss": 0.5094435214996338, "lr": 3e-05, "grad_norm": 0.33234581361357024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135408062366, "tokens_per_second": 138679.39196748857, "data_wait_s": 1.405299553880468e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2a68d43c547debbae8b9fee6072651f12ee0adba921dc93ba7e294347b4ccf64
-size 853629
+oid sha256:90ed34b993572e7e8c0502d21f59fab4836278b200a307b2a6d46fdfd2331c8f
+size 855747