Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75636.41396792672,
-      "required_pages_per_second_upper_estimate": 0.006555972664001282
+      "required_token_positions_per_second": 75630.60254063262,
+      "required_pages_per_second_upper_estimate": 0.006555468944212357
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9454.55174599084,
-      "required_pages_per_second_upper_estimate": 0.003771562049621366
+      "required_token_positions_per_second": 9453.825317579078,
+      "required_pages_per_second_upper_estimate": 0.003771272266466841
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18909.10349198168,
-      "required_pages_per_second_upper_estimate": 0.004374288304014654
+      "required_token_positions_per_second": 18907.650635158156,
+      "required_pages_per_second_upper_estimate": 0.004373952211157938
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28363.65523797252,
-      "required_pages_per_second_upper_estimate": 0.0017271977913943365
+      "required_token_positions_per_second": 28361.475952737233,
+      "required_pages_per_second_upper_estimate": 0.0017270650843573335
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28363.65523797252,
-      "required_pages_per_second_upper_estimate": 0.002183512695842047
+      "required_token_positions_per_second": 28361.475952737233,
+      "required_pages_per_second_upper_estimate": 0.002183344928431987
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28363.65523797252,
-      "required_pages_per_second_upper_estimate": 0.35218167100801523
+      "required_token_positions_per_second": 28361.475952737233,
+      "required_pages_per_second_upper_estimate": 0.3521546115789914
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08550059055477277,
-    "producer_p95_seconds": 0.22137790600027074,
-    "gpu_update_seconds": 0.9747685821179204,
-    "producer_batches_per_second": 11.695825648822709,
-    "gpu_batches_per_second": 1.0258845210493532,
-    "producer_headroom": 11.400723384401319,
+    "producer_batch_seconds": 0.0792378665312583,
+    "producer_p95_seconds": 0.2171078139999736,
+    "gpu_update_seconds": 0.9748434829722473,
+    "producer_batches_per_second": 12.620228733764721,
+    "gpu_batches_per_second": 1.0258056985220354,
+    "producer_headroom": 12.302747734729635,
     "sustainable": true,
-    "buffer_seconds": 60.43565209131106,
+    "buffer_seconds": 60.440295944279335,
     "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": 434795.3200518696,
-    "upload_bytes_per_second": 4198238.345037328,
+    "generation_bytes_per_second": 523269.04561499885,
+    "upload_bytes_per_second": 4017522.9884006456,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.29695416609827,
+    "estimated_full_shards_per_hour": 112.28135610902285,
     "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": 1.405440059999819,
-      "bytes": 15806866,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.6009747379994224,
       "bytes": 7396060,
@@ -562,6 +556,12 @@
       "bytes": 7512102,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 2.3653303009996307,
+      "bytes": 1160564,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -147,3 +147,4 @@
 {"step": 329700, "loss": 0.5373153388500214, "lr": 0.0003, "grad_norm": 0.15336776926794268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48259548074, "tokens_per_second": 147798.01532501896, "data_wait_s": 1.5064999388414435e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 329800, "loss": 0.5932347755879164, "lr": 0.0003, "grad_norm": 0.15959028214622506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48274110536, "tokens_per_second": 148405.0384085669, "data_wait_s": 1.3793000107398257e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 329900, "loss": 0.5818005800247192, "lr": 0.0003, "grad_norm": 0.16854894922756233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48288740098, "tokens_per_second": 148915.43229986343, "data_wait_s": 1.1409001672291197e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 330000, "loss": 0.5456385742872953, "lr": 0.0003, "grad_norm": 0.14405760445189703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48303357762, "tokens_per_second": 148276.12502650137, "data_wait_s": 1.3793998732580803e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50ebf408f3e01bb70026b30244a6fe1b888c910aca6561e776da101a4192d71a
-size 317313
+oid sha256:bae84b6a79e2a864e15df387a1e49e1c5576cd41e988fdd258e06459f6a1f5ae
+size 319431