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": 77006.61303467462,
-      "required_pages_per_second_upper_estimate": 0.005327496169698325
+      "required_token_positions_per_second": 76670.45861072722,
+      "required_pages_per_second_upper_estimate": 0.005304240226663919
     },
     {
       "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": 9625.826629334328,
-      "required_pages_per_second_upper_estimate": 0.00403013259958464
+      "required_token_positions_per_second": 9583.807326340902,
+      "required_pages_per_second_upper_estimate": 0.004012539994884119
     },
     {
       "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": 19251.653258668655,
-      "required_pages_per_second_upper_estimate": 0.004009611673542888
+      "required_token_positions_per_second": 19167.614652681805,
+      "required_pages_per_second_upper_estimate": 0.003992108648162903
     },
     {
       "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": 28877.47988800298,
-      "required_pages_per_second_upper_estimate": 0.001731160825927078
+      "required_token_positions_per_second": 28751.421979022707,
+      "required_pages_per_second_upper_estimate": 0.0017236038467629926
     },
     {
       "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": 28877.47988800298,
-      "required_pages_per_second_upper_estimate": 0.002245097042681945
+      "required_token_positions_per_second": 28751.421979022707,
+      "required_pages_per_second_upper_estimate": 0.002235296594728872
     },
     {
       "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": 28877.47988800298,
-      "required_pages_per_second_upper_estimate": 0.3704093057811339
+      "required_token_positions_per_second": 28751.421979022707,
+      "required_pages_per_second_upper_estimate": 0.3687923702751723
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09469129856250902,
-    "producer_p95_seconds": 0.22700542300117377,
-    "gpu_update_seconds": 0.9574242664951603,
-    "producer_batches_per_second": 10.560632446495232,
-    "gpu_batches_per_second": 1.0444690353010337,
-    "producer_headroom": 10.111005773810687,
+    "producer_batch_seconds": 0.07792256906252248,
+    "producer_p95_seconds": 0.19477402200027427,
+    "gpu_update_seconds": 0.9616220032585597,
+    "producer_batches_per_second": 12.833252445740504,
+    "gpu_batches_per_second": 1.039909649125532,
+    "producer_headroom": 12.340737925195794,
     "sustainable": true,
-    "buffer_seconds": 60.3177287891951,
+    "buffer_seconds": 60.58218620528926,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272063.97629002266,
-    "upload_bytes_per_second": 3378570.1730552064,
+    "generation_bytes_per_second": 272112.7876064215,
+    "upload_bytes_per_second": 3445301.604025005,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.378595986609554,
+    "estimated_full_shards_per_hour": 58.389069758839454,
     "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": 1.7043199660001846,
-      "bytes": 963598,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.1770538490000035,
       "bytes": 8003790,
@@ -562,6 +556,12 @@
       "bytes": 10689415,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.1898289029995794,
+      "bytes": 7719194,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -123,3 +123,4 @@
 {"step": 697300, "loss": 0.5319822430610657, "lr": 8e-05, "grad_norm": 0.19870879359192156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102065739429, "tokens_per_second": 104200.85483742703, "data_wait_s": 1.6117999621201307e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 697400, "loss": 0.5253143310546875, "lr": 8e-05, "grad_norm": 0.21027301017552655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102080390446, "tokens_per_second": 152495.90607963104, "data_wait_s": 2.0289999156375416e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 697500, "loss": 0.5619334578514099, "lr": 8e-05, "grad_norm": 0.20341879759776943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102095144107, "tokens_per_second": 153889.78511646157, "data_wait_s": 1.642700044612866e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 697600, "loss": 0.5490195751190186, "lr": 8e-05, "grad_norm": 0.20628541522345373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102109910778, "tokens_per_second": 154338.96060391763, "data_wait_s": 1.7364000086672604e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:85923a3bd059bf40eab1b5dcaeb0b0c504149049424ca79e20bf510fcc982150
-size 266481
+oid sha256:15eb1a68f96005d7d147ea50178d99784cc3f0a17c8b388154787496b15f92c4
+size 268599