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": 76914.9504357288,
-      "required_pages_per_second_upper_estimate": 0.005321154738416721
+      "required_token_positions_per_second": 77222.45280822497,
+      "required_pages_per_second_upper_estimate": 0.005342428466049812
     },
     {
       "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": 9614.3688044661,
-      "required_pages_per_second_upper_estimate": 0.004025335447578904
+      "required_token_positions_per_second": 9652.806601028122,
+      "required_pages_per_second_upper_estimate": 0.004041428550326956
     },
     {
       "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": 19228.7376089322,
-      "required_pages_per_second_upper_estimate": 0.004004838948029107
+      "required_token_positions_per_second": 19305.613202056244,
+      "required_pages_per_second_upper_estimate": 0.004020850106731121
     },
     {
       "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": 28843.106413398298,
-      "required_pages_per_second_upper_estimate": 0.001729100188609784
+      "required_token_positions_per_second": 28958.419803084365,
+      "required_pages_per_second_upper_estimate": 0.0017360130502481128
     },
     {
       "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": 28843.106413398298,
-      "required_pages_per_second_upper_estimate": 0.002242424656224367
+      "required_token_positions_per_second": 28958.419803084365,
+      "required_pages_per_second_upper_estimate": 0.002251389765062461
     },
     {
       "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": 28843.106413398298,
-      "required_pages_per_second_upper_estimate": 0.36996839975626655
+      "required_token_positions_per_second": 28958.419803084365,
+      "required_pages_per_second_upper_estimate": 0.3714475161052881
     }
   ],
   "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": 0.11919012630448833,
-    "producer_p95_seconds": 0.3533010520040989,
-    "gpu_update_seconds": 0.9585652669906892,
-    "producer_batches_per_second": 8.389956710385189,
-    "gpu_batches_per_second": 1.0432257817346027,
-    "producer_headroom": 8.042321094130703,
+    "producer_batch_seconds": 0.09050281592965348,
+    "producer_p95_seconds": 0.31492198000341887,
+    "gpu_update_seconds": 0.9547482282529521,
+    "producer_batches_per_second": 11.049379952744072,
+    "gpu_batches_per_second": 1.047396549590725,
+    "producer_headroom": 10.54937593317609,
     "sustainable": true,
-    "buffer_seconds": 60.38961182041342,
+    "buffer_seconds": 60.149138379935984,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399662.8063795618,
-    "upload_bytes_per_second": 5712805.779824095,
+    "generation_bytes_per_second": 399644.37986343034,
+    "upload_bytes_per_second": 5528497.713898112,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.7583345750822,
+    "estimated_full_shards_per_hour": 85.75438067366774,
     "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": 1.4275493559980532,
-      "bytes": 16198066,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.0836265130019456,
       "bytes": 2657150,
@@ -562,6 +556,12 @@
       "bytes": 6049845,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1915414559989586,
+      "bytes": 1026854,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -372,3 +372,4 @@
 {"step": 722200, "loss": 0.5516434907913208, "lr": 8e-05, "grad_norm": 0.22077290950308204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105710086105, "tokens_per_second": 150220.9483061592, "data_wait_s": 1.7589998606126755e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 722300, "loss": 0.5085393190383911, "lr": 8e-05, "grad_norm": 0.22677287371391655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105724561000, "tokens_per_second": 150868.86048445228, "data_wait_s": 1.598599919816479e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 722400, "loss": 0.5172302722930908, "lr": 8e-05, "grad_norm": 0.21287814167332977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105739363686, "tokens_per_second": 154658.96207514982, "data_wait_s": 1.5737001376692206e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 722500, "loss": 0.519875705242157, "lr": 8e-05, "grad_norm": 0.22840929185045208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105754059450, "tokens_per_second": 151841.19993489745, "data_wait_s": 1.5067998901940882e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f45e04209b4a357ad504f11f608053101319991e037dd8d19dba7d9fe8f15ab1
-size 797778
+oid sha256:d95e48ff11535cdd3ac0bd58d147606c27df6d0ea2f5f48c2b10332ecef0b536
+size 799896