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": 75628.53690589264,
-      "required_pages_per_second_upper_estimate": 0.006555289900228399
+      "required_token_positions_per_second": 76169.42823155128,
+      "required_pages_per_second_upper_estimate": 0.006602172989459958
     },
     {
       "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": 9453.56711323658,
-      "required_pages_per_second_upper_estimate": 0.00377116926489412
+      "required_token_positions_per_second": 9521.17852894391,
+      "required_pages_per_second_upper_estimate": 0.003798140469500523
     },
     {
       "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": 18907.13422647316,
-      "required_pages_per_second_upper_estimate": 0.0043738327490971465
+      "required_token_positions_per_second": 19042.35705788782,
+      "required_pages_per_second_upper_estimate": 0.004405114171304377
     },
     {
       "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": 28360.701339709743,
-      "required_pages_per_second_upper_estimate": 0.0017270179144880397
+      "required_token_positions_per_second": 28563.535586831735,
+      "required_pages_per_second_upper_estimate": 0.0017393694559487193
     },
     {
       "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": 28360.701339709743,
-      "required_pages_per_second_upper_estimate": 0.0021832852965768643
+      "required_token_positions_per_second": 28563.535586831735,
+      "required_pages_per_second_upper_estimate": 0.002198900038401448
     },
     {
       "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": 28360.701339709743,
-      "required_pages_per_second_upper_estimate": 0.35214499347765305
+      "required_token_positions_per_second": 28563.535586831735,
+      "required_pages_per_second_upper_estimate": 0.3546635159843517
     }
   ],
   "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": 0.3602242562968172,
-    "producer_p95_seconds": 0.39618910999706713,
-    "gpu_update_seconds": 0.9748701087757713,
-    "producer_batches_per_second": 2.7760484823543394,
-    "gpu_batches_per_second": 1.0257776815577888,
-    "producer_headroom": 2.7062866859595895,
+    "producer_batch_seconds": 0.1438636594451168,
+    "producer_p95_seconds": 0.38428035599645227,
+    "gpu_update_seconds": 0.9679473997870974,
+    "producer_batches_per_second": 6.951025741017623,
+    "gpu_batches_per_second": 1.0331139896857542,
+    "producer_headroom": 6.72822729187119,
     "sustainable": true,
-    "buffer_seconds": 60.44194674409782,
+    "buffer_seconds": 60.01273878680004,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340768.60441605636,
-    "upload_bytes_per_second": 6104115.376714776,
+    "generation_bytes_per_second": 354119.56360140035,
+    "upload_bytes_per_second": 6027280.490812045,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.12100982056873,
+    "estimated_full_shards_per_hour": 75.98581486732013,
     "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": 1.0869366869992518,
-      "bytes": 13432151,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 0.7862899389983795,
       "bytes": 1239440,
@@ -562,6 +556,12 @@
       "bytes": 10226084,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.0265658620046452,
+      "bytes": 7737951,
+      "pages": 42,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -367,3 +367,4 @@
 {"step": 76700, "loss": 0.604017112404108, "lr": 0.0003, "grad_norm": 0.1460994252826713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11225016326, "tokens_per_second": 148055.4112062978, "data_wait_s": 1.1598000128287822e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 76800, "loss": 0.6558533720672131, "lr": 0.0003, "grad_norm": 0.14504643842668982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11239515333, "tokens_per_second": 148209.24101931142, "data_wait_s": 8.47400224301964e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 76900, "loss": 0.6268212832510471, "lr": 0.0003, "grad_norm": 0.12912131628338816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11254027747, "tokens_per_second": 149133.30097234983, "data_wait_s": 1.1714997526723891e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 77000, "loss": 0.6380148641765118, "lr": 0.0003, "grad_norm": 0.14502346248895828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11268718373, "tokens_per_second": 150871.65479264862, "data_wait_s": 8.916998922359198e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8ec8744d3a0b59ac2f23a48f9d19aa0d2e011a5f7ee2b037798897c0cd2dcc1f
-size 787188
+oid sha256:8854d98f9628e6b86e99203d8c76de6b46d81d25bd39b0c8c0123f20b8e03bda
+size 789306