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": 75668.64057958657,
-      "required_pages_per_second_upper_estimate": 0.0065587659850223565
+      "required_token_positions_per_second": 75881.75468107728,
+      "required_pages_per_second_upper_estimate": 0.006577238175206819
     },
     {
       "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": 9458.580072448322,
-      "required_pages_per_second_upper_estimate": 0.003773169009274103
+      "required_token_positions_per_second": 9485.21933513466,
+      "required_pages_per_second_upper_estimate": 0.003783795809452154
     },
     {
       "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": 18917.160144896643,
-      "required_pages_per_second_upper_estimate": 0.004376152068874282
+      "required_token_positions_per_second": 18970.43867026932,
+      "required_pages_per_second_upper_estimate": 0.004388477118049229
     },
     {
       "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": 28375.740217344966,
-      "required_pages_per_second_upper_estimate": 0.0017279337032331317
+      "required_token_positions_per_second": 28455.658005403977,
+      "required_pages_per_second_upper_estimate": 0.0017328002772296965
     },
     {
       "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": 28375.740217344966,
-      "required_pages_per_second_upper_estimate": 0.0021844430310075016
+      "required_token_positions_per_second": 28455.658005403977,
+      "required_pages_per_second_upper_estimate": 0.0021905953235588807
     },
     {
       "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": 28375.740217344966,
-      "required_pages_per_second_upper_estimate": 0.3523317260059968
+      "required_token_positions_per_second": 28455.658005403977,
+      "required_pages_per_second_upper_estimate": 0.3533240374660588
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.2155270282656545,
-    "producer_p95_seconds": 0.34355466200213414,
-    "gpu_update_seconds": 0.9743534367114015,
-    "producer_batches_per_second": 4.639789301819812,
-    "gpu_batches_per_second": 1.0263216224444793,
-    "producer_headroom": 4.520794651844928,
+    "producer_batch_seconds": 0.08528839431255619,
+    "producer_p95_seconds": 0.30072691700115683,
+    "gpu_update_seconds": 0.9716169626001762,
+    "producer_batches_per_second": 11.72492468711865,
+    "gpu_batches_per_second": 1.0292121674408268,
+    "producer_headroom": 11.392135711214044,
     "sustainable": true,
-    "buffer_seconds": 60.409913076106896,
+    "buffer_seconds": 60.24025168121092,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379870.15779903,
-    "upload_bytes_per_second": 4638074.420839191,
+    "generation_bytes_per_second": 380068.25996542524,
+    "upload_bytes_per_second": 4602688.965301513,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.51129293897796,
+    "estimated_full_shards_per_hour": 81.55380105230397,
     "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": 1.4154862130017136,
-      "bytes": 9920178,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.3325980400004482,
       "bytes": 6856977,
@@ -562,6 +556,12 @@
       "bytes": 13571269,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.1989007740012312,
+      "bytes": 5658429,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -312,3 +312,4 @@
 {"step": 186200, "loss": 0.5893564112484455, "lr": 0.0003, "grad_norm": 0.1427987720191481, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27251538034, "tokens_per_second": 150829.92931331764, "data_wait_s": 1.2885997421108186e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 186300, "loss": 0.6419283822178841, "lr": 0.0003, "grad_norm": 0.15514593585242306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27266249770, "tokens_per_second": 150714.2714817881, "data_wait_s": 8.15699968370609e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 186400, "loss": 0.5877294652163982, "lr": 0.0003, "grad_norm": 0.14422305458464849, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27280927819, "tokens_per_second": 150316.6404611747, "data_wait_s": 1.4843000826658681e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 186500, "loss": 0.6012108605355024, "lr": 0.0003, "grad_norm": 0.15134514336454125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27295665567, "tokens_per_second": 150954.87258442675, "data_wait_s": 1.92320003407076e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eadbb7ffbb7a122a035d60f846b7b981a505e27d80f7dd3095adda997df10abb
-size 669915
+oid sha256:70f5e8a514186cbcce876e5ebd2f6fc45fa4759a7e7096ce469c7786537afc1c
+size 672033