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": 76872.30517150898,
-      "required_pages_per_second_upper_estimate": 0.0053182044400873485
+      "required_token_positions_per_second": 76299.85449539992,
+      "required_pages_per_second_upper_estimate": 0.005278600974045556
     },
     {
       "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": 9609.038146438623,
-      "required_pages_per_second_upper_estimate": 0.004023103612379598
+      "required_token_positions_per_second": 9537.48181192499,
+      "required_pages_per_second_upper_estimate": 0.003993144469384923
     },
     {
       "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": 19218.076292877246,
-      "required_pages_per_second_upper_estimate": 0.004002618477052671
+      "required_token_positions_per_second": 19074.96362384998,
+      "required_pages_per_second_upper_estimate": 0.003972811882229174
     },
     {
       "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": 28827.11443931587,
-      "required_pages_per_second_upper_estimate": 0.0017281414941818728
+      "required_token_positions_per_second": 28612.445435774975,
+      "required_pages_per_second_upper_estimate": 0.0017152724152001855
     },
     {
       "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": 28827.11443931587,
-      "required_pages_per_second_upper_estimate": 0.002241181350580718
+      "required_token_positions_per_second": 28612.445435774975,
+      "required_pages_per_second_upper_estimate": 0.002224491779784571
     },
     {
       "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": 28827.11443931587,
-      "required_pages_per_second_upper_estimate": 0.3697632718835811
+      "required_token_positions_per_second": 28612.445435774975,
+      "required_pages_per_second_upper_estimate": 0.36700972839977647
     }
   ],
   "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": 0.1094828718124603,
-    "producer_p95_seconds": 0.20776308000131394,
-    "gpu_update_seconds": 0.9590970354733899,
-    "producer_batches_per_second": 9.133848824434924,
-    "gpu_batches_per_second": 1.0426473683201631,
-    "producer_headroom": 8.760247329977643,
+    "producer_batch_seconds": 0.09201112900002784,
+    "producer_p95_seconds": 0.20955274500011,
+    "gpu_update_seconds": 0.966292799476374,
+    "producer_batches_per_second": 10.868250513475358,
+    "gpu_batches_per_second": 1.0348830091064443,
+    "producer_headroom": 10.501912214076643,
     "sustainable": true,
-    "buffer_seconds": 60.42311323482356,
+    "buffer_seconds": 60.87644636701157,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 173,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 501147.21180226485,
-    "upload_bytes_per_second": 2178019.4106091666,
+    "generation_bytes_per_second": 631499.0001856502,
+    "upload_bytes_per_second": 2163042.7768235332,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.5345255427452,
+    "estimated_full_shards_per_hour": 67.75249244774403,
     "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": 1.6013349519998883,
-      "bytes": 17762321,
-      "pages": 31,
-      "shards": 5
-    },
     {
       "seconds": 1.5430965340001421,
       "bytes": 6138495,
@@ -562,6 +556,12 @@
       "bytes": 6934679,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1427723569995578,
+      "bytes": 13775084,
+      "pages": 72,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -137,3 +137,4 @@
 {"step": 658700, "loss": 0.5515803694725037, "lr": 0.0001, "grad_norm": 0.1961138554767908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96415753876, "tokens_per_second": 153168.6945994227, "data_wait_s": 8.998998964671046e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 658800, "loss": 0.5704889893531799, "lr": 0.0001, "grad_norm": 0.19855481073692224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96430477585, "tokens_per_second": 153797.17199487513, "data_wait_s": 1.0458999895490706e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 658900, "loss": 0.5716390013694763, "lr": 0.0001, "grad_norm": 0.19920145486725915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96445199974, "tokens_per_second": 153532.37566865486, "data_wait_s": 1.3941998986410908e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 659000, "loss": 0.5488200783729553, "lr": 0.0001, "grad_norm": 0.17847421259339394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96459949323, "tokens_per_second": 154457.25089723495, "data_wait_s": 1.0686000678106211e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d81e1ce413ae35553e694ed5fb2273f0b89c16894f89fcc1c3542cc0a50e517
-size 296133
+oid sha256:044b8674248b215ffe78f4645b6001c6b0f025c6eb2c880c011c3ac857913a23
+size 298251