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": 77111.78933258702,
-      "required_pages_per_second_upper_estimate": 0.005334772510030538
+      "required_token_positions_per_second": 76736.30222281809,
+      "required_pages_per_second_upper_estimate": 0.005308795440526594
     },
     {
       "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": 9638.973666573378,
-      "required_pages_per_second_upper_estimate": 0.004035636989535274
+      "required_token_positions_per_second": 9592.037777852262,
+      "required_pages_per_second_upper_estimate": 0.004015985913060553
     },
     {
       "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": 19277.947333146756,
-      "required_pages_per_second_upper_estimate": 0.004015088035835302
+      "required_token_positions_per_second": 19184.075555704523,
+      "required_pages_per_second_upper_estimate": 0.00399553702015933
     },
     {
       "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": 28916.920999720132,
-      "required_pages_per_second_upper_estimate": 0.0017335252603514306
+      "required_token_positions_per_second": 28776.113333556787,
+      "required_pages_per_second_upper_estimate": 0.0017250840557657954
     },
     {
       "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": 28916.920999720132,
-      "required_pages_per_second_upper_estimate": 0.0022481634156348374
+      "required_token_positions_per_second": 28776.113333556787,
+      "required_pages_per_second_upper_estimate": 0.002237216238938106
     },
     {
       "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": 28916.920999720132,
-      "required_pages_per_second_upper_estimate": 0.3709152140136752
+      "required_token_positions_per_second": 28776.113333556787,
+      "required_pages_per_second_upper_estimate": 0.36910908445962454
     }
   ],
   "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": 0.2522882040391323,
-    "producer_p95_seconds": 0.17308578900065186,
-    "gpu_update_seconds": 0.9561183917287593,
-    "producer_batches_per_second": 3.9637207922923356,
-    "gpu_batches_per_second": 1.0458955801403405,
-    "producer_headroom": 3.789786349188392,
+    "producer_batch_seconds": 0.05418472578124067,
+    "producer_p95_seconds": 0.17214284399960889,
+    "gpu_update_seconds": 0.960796883148175,
+    "producer_batches_per_second": 18.455385453776913,
+    "gpu_batches_per_second": 1.0408027102704278,
+    "producer_headroom": 17.731876821287027,
     "sustainable": true,
-    "buffer_seconds": 60.23545867891184,
+    "buffer_seconds": 60.530203638335024,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 459908.7632964141,
-    "upload_bytes_per_second": 5520603.21527887,
+    "generation_bytes_per_second": 460014.61391499446,
+    "upload_bytes_per_second": 5452492.670439877,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.6857144753391,
+    "estimated_full_shards_per_hour": 98.70842755401016,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 13446052,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 0.8787024550001661,
+      "bytes": 837425,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -57,3 +57,4 @@
 {"step": 610100, "loss": 0.5721548199653625, "lr": 0.0001, "grad_norm": 0.14133212756874464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89301011279, "tokens_per_second": 139268.55388615286, "data_wait_s": 1.18650004878873e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 610200, "loss": 0.549430251121521, "lr": 0.0001, "grad_norm": 0.16628906229233595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89315504557, "tokens_per_second": 150681.60285144622, "data_wait_s": 1.5778000488353427e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 610300, "loss": 0.6092767119407654, "lr": 0.0001, "grad_norm": 0.14803397781210065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89330207406, "tokens_per_second": 152671.29239890774, "data_wait_s": 1.3412999578577e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 610400, "loss": 0.5435611009597778, "lr": 0.0001, "grad_norm": 0.15360692895727052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89344914056, "tokens_per_second": 152009.4576179536, "data_wait_s": 1.6773999959696084e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe2f62dc28fd6869afe3f1b434a7879ae0d1d8681b5fc80434e74c69c18be438
-size 126693
+oid sha256:8cdebeee24139cfe4f8bbe0d85634e81b8c701bb4ab25072d1fac1f073c5cb39
+size 128811