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": 76108.63377546435,
-      "required_pages_per_second_upper_estimate": 0.006596903480088482
+      "required_token_positions_per_second": 76088.74782441546,
+      "required_pages_per_second_upper_estimate": 0.006595179816252047
     },
     {
       "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": 9513.579221933043,
-      "required_pages_per_second_upper_estimate": 0.003795108992314123
+      "required_token_positions_per_second": 9511.093478051933,
+      "required_pages_per_second_upper_estimate": 0.003794117391914765
     },
     {
       "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": 19027.158443866087,
-      "required_pages_per_second_upper_estimate": 0.0044015982394369196
+      "required_token_positions_per_second": 19022.186956103866,
+      "required_pages_per_second_upper_estimate": 0.004400448173238327
     },
     {
       "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": 28540.73766579913,
-      "required_pages_per_second_upper_estimate": 0.0017379811821692812
+      "required_token_positions_per_second": 28533.280434155797,
+      "required_pages_per_second_upper_estimate": 0.0017375270758872728
     },
     {
       "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": 28540.73766579913,
-      "required_pages_per_second_upper_estimate": 0.0021971449913316733
+      "required_token_positions_per_second": 28533.280434155797,
+      "required_pages_per_second_upper_estimate": 0.002196570912996831
     },
     {
       "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": 28540.73766579913,
-      "required_pages_per_second_upper_estimate": 0.35438044210486025
+      "required_token_positions_per_second": 28533.280434155797,
+      "required_pages_per_second_upper_estimate": 0.3542878482455989
     }
   ],
   "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": 0.10494775075002849,
-    "producer_p95_seconds": 0.351025005998963,
-    "gpu_update_seconds": 0.9687205819186336,
-    "producer_batches_per_second": 9.528551044241684,
-    "gpu_batches_per_second": 1.0322894121021098,
-    "producer_headroom": 9.230503512419208,
+    "producer_batch_seconds": 0.11943702612541074,
+    "producer_p95_seconds": 0.3365304600010859,
+    "gpu_update_seconds": 0.9689737590390739,
+    "producer_batches_per_second": 8.37261301993558,
+    "gpu_batches_per_second": 1.0320196916288988,
+    "producer_headroom": 8.112842310906471,
     "sustainable": true,
-    "buffer_seconds": 60.06067607895528,
+    "buffer_seconds": 60.07637306042258,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 168,
+    "upload_pages_per_commit": 166,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 685546.3634471458,
-    "upload_bytes_per_second": 5172623.613788709,
+    "generation_bytes_per_second": 677685.2122417946,
+    "upload_bytes_per_second": 5127123.783718732,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.55114544659033,
+    "estimated_full_shards_per_hour": 72.70773542137326,
     "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": 1.2069239850025042,
-      "bytes": 7251816,
-      "pages": 27,
-      "shards": 3
-    },
     {
       "seconds": 1.169119137000962,
       "bytes": 6109087,
@@ -562,6 +556,12 @@
       "bytes": 6405900,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9840362190006999,
+      "bytes": 2390157,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -302,3 +302,4 @@
 {"step": 145200, "loss": 0.6110075414180756, "lr": 0.0003, "grad_norm": 0.13060606311302397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21249704512, "tokens_per_second": 149717.61129311027, "data_wait_s": 1.4770001143915579e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 145300, "loss": 0.610079487785697, "lr": 0.0003, "grad_norm": 0.1429218621358837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21264470418, "tokens_per_second": 151096.84181564505, "data_wait_s": 1.37290007842239e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 145400, "loss": 0.6292114295065403, "lr": 0.0003, "grad_norm": 0.1463837511759007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21279195490, "tokens_per_second": 151062.97558532076, "data_wait_s": 1.0263000149279833e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 145500, "loss": 0.574711250141263, "lr": 0.0003, "grad_norm": 0.16296621779060388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21293792544, "tokens_per_second": 148855.49993111668, "data_wait_s": 9.462997695663944e-06, "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 sha256:16c7c1a128852bf6dd918a8fc17fc24e04cf3f8e4b86b6a0c0fe8175e5865fd3
-size 648735
+oid sha256:57bf737f09e01e9f7d6680e69e91370d64cdc5b98bce5ee543323e036f5428f5
+size 650853