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": 75787.00086585313,
-      "required_pages_per_second_upper_estimate": 0.006569025154654539
+      "required_token_positions_per_second": 76145.06922605264,
+      "required_pages_per_second_upper_estimate": 0.006600061612600671
     },
     {
       "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": 9473.375108231641,
-      "required_pages_per_second_upper_estimate": 0.003779070970253567
+      "required_token_positions_per_second": 9518.13365325658,
+      "required_pages_per_second_upper_estimate": 0.0037969258230638983
     },
     {
       "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": 18946.750216463282,
-      "required_pages_per_second_upper_estimate": 0.004382997211163751
+      "required_token_positions_per_second": 19036.26730651316,
+      "required_pages_per_second_upper_estimate": 0.004403705414499813
     },
     {
       "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": 28420.125324694924,
-      "required_pages_per_second_upper_estimate": 0.001730636523400083
+      "required_token_positions_per_second": 28554.40095976974,
+      "required_pages_per_second_upper_estimate": 0.001738813205086328
     },
     {
       "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": 28420.125324694924,
-      "required_pages_per_second_upper_estimate": 0.002187859919437142
+      "required_token_positions_per_second": 28554.40095976974,
+      "required_pages_per_second_upper_estimate": 0.0021981968295239494
     },
     {
       "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": 28420.125324694924,
-      "required_pages_per_second_upper_estimate": 0.3528828404918848
+      "required_token_positions_per_second": 28554.40095976974,
+      "required_pages_per_second_upper_estimate": 0.3545500944878719
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09318208592964794,
-    "producer_p95_seconds": 0.19062702499923034,
-    "gpu_update_seconds": 0.9728317410330346,
-    "producer_batches_per_second": 10.731676480766867,
-    "gpu_batches_per_second": 1.0279269865702736,
-    "producer_headroom": 10.440115514987701,
+    "producer_batch_seconds": 0.11036811774217625,
+    "producer_p95_seconds": 0.18251931400027388,
+    "gpu_update_seconds": 0.968257048675377,
+    "producer_batches_per_second": 9.060587608606633,
+    "gpu_batches_per_second": 1.0327835995287087,
+    "producer_headroom": 8.77297781717415,
     "sustainable": true,
-    "buffer_seconds": 60.31556794404815,
+    "buffer_seconds": 60.031937017873375,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326864.67578032607,
-    "upload_bytes_per_second": 4958695.103981961,
+    "generation_bytes_per_second": 325569.43929541245,
+    "upload_bytes_per_second": 4969415.092382257,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.13755040223442,
+    "estimated_full_shards_per_hour": 69.85962280413418,
     "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": [
@@ -376,6 +376,12 @@
       "bytes": 10793699,
       "pages": 45,
       "shards": 3
+    },
+    {
+      "seconds": 1.3661847930006843,
+      "bytes": 7242166,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -68,3 +68,4 @@
 {"step": 161800, "loss": 0.5861719343811274, "lr": 0.0003, "grad_norm": 0.14012673183850052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23680373961, "tokens_per_second": 140361.65730322493, "data_wait_s": 1.0755999937828165e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 161900, "loss": 0.585586529225111, "lr": 0.0003, "grad_norm": 0.13278766848550524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23695015059, "tokens_per_second": 149090.53104419197, "data_wait_s": 1.2470999536162708e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 162000, "loss": 0.6017758883535862, "lr": 0.0003, "grad_norm": 0.13104959586342854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23709731254, "tokens_per_second": 150769.64504119346, "data_wait_s": 9.796000085771084e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 162100, "loss": 0.5705337896943092, "lr": 0.0003, "grad_norm": 0.14917666228697132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23724411562, "tokens_per_second": 150661.94945537805, "data_wait_s": 1.3358000614971388e-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 sha256:4857e3c37782b2d0d959edbbd33ad02d5f235d360881cf33250d2a75192ae1e7
-size 149208
+oid sha256:7d61f1e7146e559f1ab60b2c2090011574f95fe5da34b467fb1d476f625fce1f
+size 151326