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": 74960.78117479333,
-      "required_pages_per_second_upper_estimate": 0.0064974105258681384
+      "required_token_positions_per_second": 75368.18777639217,
+      "required_pages_per_second_upper_estimate": 0.006532723497532133
     },
     {
       "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": 9370.097646849166,
-      "required_pages_per_second_upper_estimate": 0.0037378720467724454
+      "required_token_positions_per_second": 9421.02347204902,
+      "required_pages_per_second_upper_estimate": 0.0037581871198536063
     },
     {
       "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": 18740.195293698333,
-      "required_pages_per_second_upper_estimate": 0.004335214365024561
+      "required_token_positions_per_second": 18842.04694409804,
+      "required_pages_per_second_upper_estimate": 0.004358775951816709
     },
     {
       "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": 28110.292940547497,
-      "required_pages_per_second_upper_estimate": 0.0017117693567703942
+      "required_token_positions_per_second": 28263.070416147064,
+      "required_pages_per_second_upper_estimate": 0.0017210727034729413
     },
     {
       "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": 28110.292940547497,
-      "required_pages_per_second_upper_estimate": 0.002164008164834541
+      "required_token_positions_per_second": 28263.070416147064,
+      "required_pages_per_second_upper_estimate": 0.0021757693978213155
     },
     {
       "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": 28110.292940547497,
-      "required_pages_per_second_upper_estimate": 0.3490357592230589
+      "required_token_positions_per_second": 28263.070416147064,
+      "required_pages_per_second_upper_estimate": 0.3509327441566865
     }
   ],
   "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": 0.09609352081292855,
-    "producer_p95_seconds": 0.3889136040015728,
-    "gpu_update_seconds": 0.9835543179316831,
-    "producer_batches_per_second": 10.406528884988662,
-    "gpu_batches_per_second": 1.016720664805682,
-    "producer_headroom": 10.235386419511382,
+    "producer_batch_seconds": 0.1606457544066302,
+    "producer_p95_seconds": 0.3911977770039812,
+    "gpu_update_seconds": 0.9782376646595458,
+    "producer_batches_per_second": 6.2248766155922,
+    "gpu_batches_per_second": 1.0222464704914302,
+    "producer_headroom": 6.089408763230731,
     "sustainable": true,
-    "buffer_seconds": 60.98036771176435,
+    "buffer_seconds": 60.65073520889184,
     "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": 349168.267005733,
-    "upload_bytes_per_second": 5476846.714755958,
+    "generation_bytes_per_second": 362750.0021895322,
+    "upload_bytes_per_second": 5470118.08714139,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.92338187817566,
+    "estimated_full_shards_per_hour": 77.83770608200525,
     "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": 1.0085502450019703,
-      "bytes": 1017261,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.7747774790004769,
       "bytes": 7374031,
@@ -562,6 +556,12 @@
       "bytes": 6119324,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0826924209977733,
+      "bytes": 902486,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 149700, "loss": 0.6227236445993185, "lr": 0.0003, "grad_norm": 0.17610612671842374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21908181421, "tokens_per_second": 149790.0730665227, "data_wait_s": 1.3098004274070263e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 149800, "loss": 0.662655808031559, "lr": 0.0003, "grad_norm": 0.16492261881099443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21922915407, "tokens_per_second": 151305.0044205704, "data_wait_s": 1.0435003787279129e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 149900, "loss": 0.6620606575161219, "lr": 0.0003, "grad_norm": 0.15283104603146141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21937642348, "tokens_per_second": 149515.8032362199, "data_wait_s": 1.1324002116452903e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 150000, "loss": 0.6477003693580627, "lr": 0.0003, "grad_norm": 0.14979309409476083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21952273112, "tokens_per_second": 148775.43239117687, "data_wait_s": 1.6620004316791892e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eec0650a9d3254e4f3d1cca442136d547f97f3d7fa9246d0215c7f540a5d96da
-size 744045
+oid sha256:959e634aa53f4e3242414aa8022d4fa2cd616f24ec4853cb8c855fe5b39c67a8
+size 746163