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": 75778.54940363606,
-      "required_pages_per_second_upper_estimate": 0.005242535878202869
+      "required_token_positions_per_second": 75678.17466533955,
+      "required_pages_per_second_upper_estimate": 0.005235591720906036
     },
     {
       "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": 9472.318675454508,
-      "required_pages_per_second_upper_estimate": 0.0039658620249057586
+      "required_token_positions_per_second": 9459.771833167444,
+      "required_pages_per_second_upper_estimate": 0.003960608924048026
     },
     {
       "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": 18944.637350909015,
-      "required_pages_per_second_upper_estimate": 0.0039456683565105115
+      "required_token_positions_per_second": 18919.54366633489,
+      "required_pages_per_second_upper_estimate": 0.003940442003778685
     },
     {
       "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": 28416.95602636352,
-      "required_pages_per_second_upper_estimate": 0.0017035531236010138
+      "required_token_positions_per_second": 28379.315499502336,
+      "required_pages_per_second_upper_estimate": 0.0017012966314894423
     },
     {
       "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": 28416.95602636352,
-      "required_pages_per_second_upper_estimate": 0.002209293338069869
+      "required_token_positions_per_second": 28379.315499502336,
+      "required_pages_per_second_upper_estimate": 0.0022063669526695925
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28416.95602636352,
-      "required_pages_per_second_upper_estimate": 0.3645022001560206
+      "required_token_positions_per_second": 28379.315499502336,
+      "required_pages_per_second_upper_estimate": 0.3640193878926942
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08540032384422602,
-    "producer_p95_seconds": 0.2961677140046959,
-    "gpu_update_seconds": 0.9729402394243025,
-    "producer_batches_per_second": 11.709557469876161,
-    "gpu_batches_per_second": 1.0278123562776158,
-    "producer_headroom": 11.392699648293942,
+    "producer_batch_seconds": 0.2663115038986348,
+    "producer_p95_seconds": 0.3241469569984474,
+    "gpu_update_seconds": 0.9742306857431019,
+    "producer_batches_per_second": 3.7550011372419965,
+    "gpu_batches_per_second": 1.0264509367586203,
+    "producer_headroom": 3.6582373329013977,
     "sustainable": true,
-    "buffer_seconds": 60.322294844306754,
+    "buffer_seconds": 60.40230251607232,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 462442.41354963457,
-    "upload_bytes_per_second": 4892056.1450199075,
+    "generation_bytes_per_second": 450734.8854318856,
+    "upload_bytes_per_second": 4897340.299530888,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.22937683932092,
+    "estimated_full_shards_per_hour": 96.71721384255815,
     "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": 1.0239220590010518,
-      "bytes": 7655031,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.4461447649991896,
       "bytes": 15647853,
@@ -562,6 +556,12 @@
       "bytes": 6097373,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.580753474998346,
+      "bytes": 10855257,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -336,3 +336,4 @@
 {"step": 793600, "loss": 0.5755289793014526, "lr": 3e-05, "grad_norm": 0.3006565562961368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116160663144, "tokens_per_second": 152079.74165767743, "data_wait_s": 1.105800038203597e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 793700, "loss": 0.5079626441001892, "lr": 3e-05, "grad_norm": 0.2647570273944449, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116175177861, "tokens_per_second": 151070.3173296538, "data_wait_s": 1.0025003575719893e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 793800, "loss": 0.5176961421966553, "lr": 3e-05, "grad_norm": 0.26464572361485117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116189747580, "tokens_per_second": 150848.97817983935, "data_wait_s": 6.815003871452063e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 793900, "loss": 0.498049259185791, "lr": 3e-05, "grad_norm": 0.28876484965245436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116204357103, "tokens_per_second": 151941.34767091012, "data_wait_s": 1.3454999134410173e-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 sha256:7fc0203d4016ee3f982946bd72b4ab21d745842d87a17ed0971626f833c821c4
-size 720747
+oid sha256:0eeaefd5d015bebb358d10ccd5510c8370234119c434398d094d078190dd9750
+size 722865