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": 76462.3341630539,
-      "required_pages_per_second_upper_estimate": 0.005289841694458628
+      "required_token_positions_per_second": 77267.37172767709,
+      "required_pages_per_second_upper_estimate": 0.005345536061123762
     },
     {
       "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": 9557.791770381738,
-      "required_pages_per_second_upper_estimate": 0.004001647824870602
+      "required_token_positions_per_second": 9658.421465959636,
+      "required_pages_per_second_upper_estimate": 0.004043779377022068
     },
     {
       "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": 19115.583540763477,
-      "required_pages_per_second_upper_estimate": 0.00398127193970301
+      "required_token_positions_per_second": 19316.842931919273,
+      "required_pages_per_second_upper_estimate": 0.004023188963313698
     },
     {
       "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": 28673.375311145217,
-      "required_pages_per_second_upper_estimate": 0.0017189250681940989
+      "required_token_positions_per_second": 28975.264397878913,
+      "required_pages_per_second_upper_estimate": 0.0017370228579858388
     },
     {
       "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": 28673.375311145217,
-      "required_pages_per_second_upper_estimate": 0.002229228809592409
+      "required_token_positions_per_second": 28975.264397878913,
+      "required_pages_per_second_upper_estimate": 0.002252699358216195
     },
     {
       "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": 28673.375311145217,
-      "required_pages_per_second_upper_estimate": 0.36779127141962287
+      "required_token_positions_per_second": 28975.264397878913,
+      "required_pages_per_second_upper_estimate": 0.3716635804809958
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.18578195855469737,
-    "producer_p95_seconds": 0.14507283600005394,
-    "gpu_update_seconds": 0.9642394625669808,
-    "producer_batches_per_second": 5.382653987392339,
-    "gpu_batches_per_second": 1.0370867806403796,
-    "producer_headroom": 5.190167387987205,
+    "producer_batch_seconds": 0.057159645890600075,
+    "producer_p95_seconds": 0.14043619400035823,
+    "gpu_update_seconds": 0.9541931911421636,
+    "producer_batches_per_second": 17.494859956164465,
+    "gpu_batches_per_second": 1.048005801427912,
+    "producer_headroom": 16.693476250157826,
     "sustainable": true,
-    "buffer_seconds": 60.74708614171979,
+    "buffer_seconds": 60.114171041956304,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 72,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397164.03355565097,
-    "upload_bytes_per_second": 3753464.9340458014,
+    "generation_bytes_per_second": 457648.17939915234,
+    "upload_bytes_per_second": 4442732.476434205,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.22215609552524,
+    "estimated_full_shards_per_hour": 98.20064579468658,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 7677212,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.560628134000126,
+      "bytes": 10720223,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -16,3 +16,4 @@
 {"step": 886600, "loss": 0.5315039753913879, "lr": 3e-05, "grad_norm": 0.31062201203068485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129773374665, "tokens_per_second": 151270.22340725388, "data_wait_s": 1.1365000091245747e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 886700, "loss": 0.5291098952293396, "lr": 3e-05, "grad_norm": 0.39168456901119597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129787989895, "tokens_per_second": 151551.92030538927, "data_wait_s": 9.362000128021464e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 886800, "loss": 0.5167833566665649, "lr": 3e-05, "grad_norm": 0.327532679284899, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129802695698, "tokens_per_second": 153200.41769600208, "data_wait_s": 1.7203999959747307e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 886900, "loss": 0.5570690035820007, "lr": 3e-05, "grad_norm": 0.31843416726930274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129817509329, "tokens_per_second": 153847.58302292152, "data_wait_s": 1.6211999991355697e-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:e120a74965270854360ceb0bcdbb6b821458c4c5e89ce3e962c2caedf0b39d14
-size 38289
+oid sha256:d0b837ab3d8ec170dc3503a680bf596a4019e992cc4a553bf48faaeec33ea0a7
+size 40407