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": 76724.70985344138,
-      "required_pages_per_second_upper_estimate": 0.005307993453515111
+      "required_token_positions_per_second": 77062.23086525087,
+      "required_pages_per_second_upper_estimate": 0.005331343940268737
     },
     {
       "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": 9590.588731680173,
-      "required_pages_per_second_upper_estimate": 0.004015379227687825
+      "required_token_positions_per_second": 9632.778858156358,
+      "required_pages_per_second_upper_estimate": 0.004033043352613378
     },
     {
       "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": 19181.177463360345,
-      "required_pages_per_second_upper_estimate": 0.003994933423951873
+      "required_token_positions_per_second": 19265.557716312716,
+      "required_pages_per_second_upper_estimate": 0.004012507605384939
     },
     {
       "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": 28771.766195040516,
-      "required_pages_per_second_upper_estimate": 0.001724823451449441
+      "required_token_positions_per_second": 28898.33657446908,
+      "required_pages_per_second_upper_estimate": 0.0017324111524344042
     },
     {
       "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": 28771.766195040516,
-      "required_pages_per_second_upper_estimate": 0.0022368782680395076
+      "required_token_positions_per_second": 28898.33657446908,
+      "required_pages_per_second_upper_estimate": 0.0022467185583158154
     },
     {
       "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": 28771.766195040516,
-      "required_pages_per_second_upper_estimate": 0.3690533240343315
+      "required_token_positions_per_second": 28898.33657446908,
+      "required_pages_per_second_upper_estimate": 0.3706768329609558
     }
   ],
   "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": 0.0503566939296789,
-    "producer_p95_seconds": 0.1615055449992724,
-    "gpu_update_seconds": 0.9609420503620586,
-    "producer_batches_per_second": 19.85833306285873,
-    "gpu_batches_per_second": 1.0406454786979353,
-    "producer_headroom": 19.08270729019613,
+    "producer_batch_seconds": 0.06655832983594223,
+    "producer_p95_seconds": 0.17118939500051056,
+    "gpu_update_seconds": 0.9567332683233499,
+    "producer_batches_per_second": 15.024415463321752,
+    "gpu_batches_per_second": 1.045223400407591,
+    "producer_headroom": 14.374358110871698,
     "sustainable": true,
-    "buffer_seconds": 60.53934917280969,
+    "buffer_seconds": 60.27419590437105,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 519161.8311701396,
-    "upload_bytes_per_second": 2331141.273349661,
+    "generation_bytes_per_second": 518637.1963744702,
+    "upload_bytes_per_second": 2414371.3091268763,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.40004350021496,
+    "estimated_full_shards_per_hour": 111.2874690859373,
     "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": [
@@ -274,6 +274,12 @@
       "bytes": 6414179,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1066383229999701,
+      "bytes": 6242311,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -60,3 +60,4 @@
 {"step": 691000, "loss": 0.5859728455543518, "lr": 8e-05, "grad_norm": 0.19990079891753568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101143788337, "tokens_per_second": 153881.91135526047, "data_wait_s": 1.6629999663564377e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 691100, "loss": 0.5225908756256104, "lr": 8e-05, "grad_norm": 0.19654221319883927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101158552772, "tokens_per_second": 153460.82393507086, "data_wait_s": 1.6150999726960436e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 691200, "loss": 0.569115936756134, "lr": 8e-05, "grad_norm": 0.2220169810367527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101173351155, "tokens_per_second": 154404.4202703566, "data_wait_s": 1.8726000234892126e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 691300, "loss": 0.5454184412956238, "lr": 8e-05, "grad_norm": 0.2134610950239632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101187872745, "tokens_per_second": 151716.51916928403, "data_wait_s": 1.7528999705973547e-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 sha256:140d981f8bda4fdaaa5c699749a59c2a9d37ef549b375b32054fd565cd090109
-size 132264
+oid sha256:84059332720d69c931c3a1da2ff517df192debd317b03b2c8fa2a86432dc6987
+size 134382