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": 76548.24674507133,
-      "required_pages_per_second_upper_estimate": 0.005295785326227248
+      "required_token_positions_per_second": 75997.07116198147,
+      "required_pages_per_second_upper_estimate": 0.00525765372048554
     },
     {
       "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": 9568.530843133916,
-      "required_pages_per_second_upper_estimate": 0.004006144050374599
+      "required_token_positions_per_second": 9499.633895247684,
+      "required_pages_per_second_upper_estimate": 0.003977298337026509
     },
     {
       "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": 19137.061686267833,
-      "required_pages_per_second_upper_estimate": 0.00398574527099478
+      "required_token_positions_per_second": 18999.267790495367,
+      "required_pages_per_second_upper_estimate": 0.003957046436374774
     },
     {
       "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": 28705.59252940175,
-      "required_pages_per_second_upper_estimate": 0.0017208564412357341
+      "required_token_positions_per_second": 28498.90168574305,
+      "required_pages_per_second_upper_estimate": 0.0017084656407570396
     },
     {
       "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": 28705.59252940175,
-      "required_pages_per_second_upper_estimate": 0.002231733556603288
+      "required_token_positions_per_second": 28498.90168574305,
+      "required_pages_per_second_upper_estimate": 0.0022156642526457646
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28705.59252940175,
-      "required_pages_per_second_upper_estimate": 0.3682045193032638
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28498.90168574305,
+      "required_pages_per_second_upper_estimate": 0.36555331108814726
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14286639046906657,
-    "producer_p95_seconds": 0.3325438909960212,
-    "gpu_update_seconds": 0.963157265319693,
-    "producer_batches_per_second": 6.999546896346625,
-    "gpu_batches_per_second": 1.0382520446108852,
-    "producer_headroom": 6.741664447162161,
+    "producer_batch_seconds": 0.09810324232046241,
+    "producer_p95_seconds": 0.33503016600297997,
+    "gpu_update_seconds": 0.9701426498773205,
+    "producer_batches_per_second": 10.193343016466436,
+    "gpu_batches_per_second": 1.030776247314202,
+    "producer_headroom": 9.888996805103227,
     "sustainable": true,
-    "buffer_seconds": 60.67890771514066,
+    "buffer_seconds": 60.14884429239387,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 527803.8889270031,
-    "upload_bytes_per_second": 4464953.446696628,
+    "generation_bytes_per_second": 509110.3901040032,
+    "upload_bytes_per_second": 4495806.40523601,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.2544279180295,
+    "estimated_full_shards_per_hour": 109.24323823299477,
     "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": 1.27796818799834,
-      "bytes": 2573475,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.1166499749961076,
       "bytes": 6085922,
@@ -562,6 +556,12 @@
       "bytes": 6724206,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.3520787300003576,
+      "bytes": 5803887,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -423,3 +423,4 @@
 {"step": 762300, "loss": 0.564041793346405, "lr": 5e-05, "grad_norm": 0.28050232875361786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111578230490, "tokens_per_second": 150239.1350671474, "data_wait_s": 1.6789999790489674e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 762400, "loss": 0.5279946327209473, "lr": 5e-05, "grad_norm": 0.2426357643554674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111592798188, "tokens_per_second": 150113.57862764425, "data_wait_s": 1.3348006177693605e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 762500, "loss": 0.5588604211807251, "lr": 5e-05, "grad_norm": 0.2513745749405481, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111607380573, "tokens_per_second": 150719.39320824147, "data_wait_s": 9.732000762596726e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 762600, "loss": 0.5727044939994812, "lr": 5e-05, "grad_norm": 0.2565716037712041, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111621971646, "tokens_per_second": 151604.20425695862, "data_wait_s": 1.292699744226411e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d9cb7233b65acb190e1261bccbc09364d9a67f33cb482c4969d4b191733d002
-size 906579
+oid sha256:3bfe9abf5bb4c458776414547c17f1cde3370118cc2e550b7e8f5a13caf19d4e
+size 908697