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": 76671.52628173959,
-      "required_pages_per_second_upper_estimate": 0.005304314090621893
+      "required_token_positions_per_second": 76200.46198615822,
+      "required_pages_per_second_upper_estimate": 0.0052717247696338196
     },
     {
       "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": 9583.940785217448,
-      "required_pages_per_second_upper_estimate": 0.00401259587132879
+      "required_token_positions_per_second": 9525.057748269777,
+      "required_pages_per_second_upper_estimate": 0.003987942773376436
     },
     {
       "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": 19167.881570434896,
-      "required_pages_per_second_upper_estimate": 0.0039921642400917775
+      "required_token_positions_per_second": 19050.115496539554,
+      "required_pages_per_second_upper_estimate": 0.003967636672599595
     },
     {
       "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": 28751.822355652344,
-      "required_pages_per_second_upper_estimate": 0.0017236278487305998
+      "required_token_positions_per_second": 28575.17324480933,
+      "required_pages_per_second_upper_estimate": 0.0017130380042631357
     },
     {
       "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": 28751.822355652344,
-      "required_pages_per_second_upper_estimate": 0.002235327722250756
+      "required_token_positions_per_second": 28575.17324480933,
+      "required_pages_per_second_upper_estimate": 0.0022215940308800346
     },
     {
       "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": 28751.822355652344,
-      "required_pages_per_second_upper_estimate": 0.36879750587668636
+      "required_token_positions_per_second": 28575.17324480933,
+      "required_pages_per_second_upper_estimate": 0.3665316407538299
     }
   ],
   "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": 0.12235711798439297,
-    "producer_p95_seconds": 0.2455014619990834,
-    "gpu_update_seconds": 0.9616086124212108,
-    "producer_batches_per_second": 8.172797925230253,
-    "gpu_batches_per_second": 1.039924130340435,
-    "producer_headroom": 7.859032872479613,
+    "producer_batch_seconds": 0.07195333206232135,
+    "producer_p95_seconds": 0.25825417599844513,
+    "gpu_update_seconds": 0.9675531890265003,
+    "producer_batches_per_second": 13.89789703045113,
+    "gpu_batches_per_second": 1.033534911921634,
+    "producer_headroom": 13.44695459257492,
     "sustainable": true,
-    "buffer_seconds": 60.58134258253628,
+    "buffer_seconds": 60.95585090866952,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382905.6804936913,
-    "upload_bytes_per_second": 5018808.802502223,
+    "generation_bytes_per_second": 386600.72028563713,
+    "upload_bytes_per_second": 4890181.098067707,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.16264544590048,
+    "estimated_full_shards_per_hour": 82.95551496912799,
     "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": 1.4052468339996267,
-      "bytes": 11659745,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.587534140999196,
       "bytes": 2763901,
@@ -562,6 +556,12 @@
       "bytes": 2637875,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 2.519623430998763,
+      "bytes": 6025157,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -240,3 +240,4 @@
 {"step": 784000, "loss": 0.5442437529563904, "lr": 3e-05, "grad_norm": 0.26176771027942847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114755616200, "tokens_per_second": 152848.74576709841, "data_wait_s": 1.0533000022405759e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 784100, "loss": 0.5269461870193481, "lr": 3e-05, "grad_norm": 0.2546602007876645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114770183930, "tokens_per_second": 151698.9817132107, "data_wait_s": 1.3642002159031108e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 784200, "loss": 0.5184525847434998, "lr": 3e-05, "grad_norm": 0.26008003421544823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114784680317, "tokens_per_second": 149795.011552634, "data_wait_s": 1.4003999240230769e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 784300, "loss": 0.4945064187049866, "lr": 3e-05, "grad_norm": 0.25559666003466647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114799206308, "tokens_per_second": 151253.90197177566, "data_wait_s": 1.1760999768739566e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9025db645dbcc45365e8a272449664c63318fd0a209aae28600ee767242aba55
-size 515853
+oid sha256:d02ba4c031b6de62317ab74cbaa65c11bf26098739860a9a7647d92b898fb9fa
+size 517971