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": 77237.35924495335,
-      "required_pages_per_second_upper_estimate": 0.005343459727930379
+      "required_token_positions_per_second": 76394.97984732335,
+      "required_pages_per_second_upper_estimate": 0.005285181966612844
     },
     {
       "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": 9654.669905619168,
-      "required_pages_per_second_upper_estimate": 0.004042208677049002
+      "required_token_positions_per_second": 9549.37248091542,
+      "required_pages_per_second_upper_estimate": 0.003998122844185811
     },
     {
       "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": 19309.339811238337,
-      "required_pages_per_second_upper_estimate": 0.004021626261146458
+      "required_token_positions_per_second": 19098.74496183084,
+      "required_pages_per_second_upper_estimate": 0.003977764907774529
     },
     {
       "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": 28964.009716857505,
-      "required_pages_per_second_upper_estimate": 0.0017363481570435082
+      "required_token_positions_per_second": 28648.11744274626,
+      "required_pages_per_second_upper_estimate": 0.0017174108975501134
     },
     {
       "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": 28964.009716857505,
-      "required_pages_per_second_upper_estimate": 0.0022518243562708908
+      "required_token_positions_per_second": 28648.11744274626,
+      "required_pages_per_second_upper_estimate": 0.002227265121422012
     },
     {
       "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": 28964.009716857505,
-      "required_pages_per_second_upper_estimate": 0.37151921751718814
+      "required_token_positions_per_second": 28648.11744274626,
+      "required_pages_per_second_upper_estimate": 0.3674672906035872
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.1886939723281671,
-    "producer_p95_seconds": 0.34065296500193654,
-    "gpu_update_seconds": 0.9545639664631252,
-    "producer_batches_per_second": 5.299586349588583,
-    "gpu_batches_per_second": 1.0475987310784687,
-    "producer_headroom": 5.058794166477112,
+    "producer_batch_seconds": 0.11753149364886895,
+    "producer_p95_seconds": 0.33364110300317407,
+    "gpu_update_seconds": 0.9650895928940179,
+    "producer_batches_per_second": 8.508357793763334,
+    "gpu_batches_per_second": 1.0361732292659962,
+    "producer_headroom": 8.2113275593797,
     "sustainable": true,
-    "buffer_seconds": 60.13752988717689,
+    "buffer_seconds": 60.800644352323125,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420624.7386015606,
-    "upload_bytes_per_second": 4891584.239696868,
+    "generation_bytes_per_second": 407623.00730782445,
+    "upload_bytes_per_second": 4971989.354840504,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.2562772611152,
+    "estimated_full_shards_per_hour": 87.4664083902936,
     "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": 0.8373393530018802,
-      "bytes": 1013057,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 1.2470868550008163,
       "bytes": 11490913,
@@ -562,6 +556,12 @@
       "bytes": 7858553,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.3608536680039833,
+      "bytes": 10702231,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -397,3 +397,4 @@
 {"step": 596800, "loss": 0.5346783995628357, "lr": 0.0003, "grad_norm": 0.164191460487655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87353465319, "tokens_per_second": 149623.66422684837, "data_wait_s": 1.2577998859342188e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 596900, "loss": 0.5882331728935242, "lr": 0.0003, "grad_norm": 0.15560431670225075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87368204448, "tokens_per_second": 153336.67514633248, "data_wait_s": 1.0984003893099725e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 597000, "loss": 0.5525102615356445, "lr": 0.0003, "grad_norm": 0.17048468384412183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87382905317, "tokens_per_second": 153100.68204656694, "data_wait_s": 9.212999430019408e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 597100, "loss": 0.5419397354125977, "lr": 0.0003, "grad_norm": 0.158056373317174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87397593914, "tokens_per_second": 152742.58909844302, "data_wait_s": 1.333500404143706e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0fd23b50807f311015b733b443ecc53901f80e904005bd6dd9d404486195b949
-size 851511
+oid sha256:2dc227a02286dbb526ff38fb64cd10ae5e5176854ab147ebee6cadac3b3c3fce
+size 853629