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": 74055.57398621921,
-      "required_pages_per_second_upper_estimate": 0.0064189494609891784
+      "required_token_positions_per_second": 74489.15134222063,
+      "required_pages_per_second_upper_estimate": 0.006456530847315619
     },
     {
       "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": 9256.946748277402,
-      "required_pages_per_second_upper_estimate": 0.0036927344615754754
+      "required_token_positions_per_second": 9311.14391777758,
+      "required_pages_per_second_upper_estimate": 0.0037143545228089914
     },
     {
       "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": 18513.893496554803,
-      "required_pages_per_second_upper_estimate": 0.0042828634270310065
+      "required_token_positions_per_second": 18622.28783555516,
+      "required_pages_per_second_upper_estimate": 0.004307938549683528
     },
     {
       "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": 27770.840244832205,
-      "required_pages_per_second_upper_estimate": 0.0016910984685719866
+      "required_token_positions_per_second": 27933.431753332738,
+      "required_pages_per_second_upper_estimate": 0.0017009994383879513
     },
     {
       "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": 27770.840244832205,
-      "required_pages_per_second_upper_estimate": 0.0021378761566531747
+      "required_token_positions_per_second": 27933.431753332738,
+      "required_pages_per_second_upper_estimate": 0.0021503929010596476
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27770.840244832205,
-      "required_pages_per_second_upper_estimate": 0.344820892817366
+      "required_token_positions_per_second": 27933.431753332738,
+      "required_pages_per_second_upper_estimate": 0.3468397351941684
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.0887014084765525,
-    "producer_p95_seconds": 0.17151205600021058,
-    "gpu_update_seconds": 0.9955766464482448,
-    "producer_batches_per_second": 11.273778141463694,
-    "gpu_batches_per_second": 1.0044430065405165,
-    "producer_headroom": 11.22391023487995,
+    "producer_batch_seconds": 0.04968172396869619,
+    "producer_p95_seconds": 0.17081841199978953,
+    "gpu_update_seconds": 0.9897817154779529,
+    "producer_batches_per_second": 20.12812600122506,
+    "gpu_batches_per_second": 1.0103237758005186,
+    "producer_headroom": 19.922451082848927,
     "sustainable": true,
-    "buffer_seconds": 60.730175433342936,
+    "buffer_seconds": 60.37668464415513,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 474408.99000670965,
-    "upload_bytes_per_second": 4629478.61418072,
+    "generation_bytes_per_second": 495447.6125568559,
+    "upload_bytes_per_second": 4562861.749715647,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.79712557936637,
+    "estimated_full_shards_per_hour": 106.31152422456033,
     "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": [
@@ -364,6 +364,12 @@
       "bytes": 7512513,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.866141241999685,
+      "bytes": 1214946,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -35,3 +35,4 @@
 {"step": 240600, "loss": 0.5745463036000729, "lr": 0.0003, "grad_norm": 0.13907632765030298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35216349372, "tokens_per_second": 147902.87925055713, "data_wait_s": 1.183400036097737e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 240700, "loss": 0.6726024448871613, "lr": 0.0003, "grad_norm": 0.18383537696960006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35231027010, "tokens_per_second": 149761.4065410573, "data_wait_s": 1.1454000741650816e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 240800, "loss": 0.6559578068554401, "lr": 0.0003, "grad_norm": 0.15070737431965414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35245699340, "tokens_per_second": 149279.59849693466, "data_wait_s": 8.551000064471737e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 240900, "loss": 0.6032705567777157, "lr": 0.0003, "grad_norm": 0.15885840097464105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35260320325, "tokens_per_second": 148073.18670679434, "data_wait_s": 1.797500044631306e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7b081fde100e9d605421d90a4ae4a7678c9c9dba9383db8435bbc8e2e0539647
-size 79314
+oid sha256:1b77cec3943b9c851eac78c657b8f559dc05a93cdd9c5a533d05e2048cb320e7
+size 81432