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": 75445.58651680796,
-      "required_pages_per_second_upper_estimate": 0.006539432224186061
+      "required_token_positions_per_second": 75725.11206895873,
+      "required_pages_per_second_upper_estimate": 0.0065636607905954535
     },
     {
       "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": 9430.698314600995,
-      "required_pages_per_second_upper_estimate": 0.003762046559199376
+      "required_token_positions_per_second": 9465.63900861984,
+      "required_pages_per_second_upper_estimate": 0.003775984924453423
     },
     {
       "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": 18861.39662920199,
-      "required_pages_per_second_upper_estimate": 0.004363252160922681
+      "required_token_positions_per_second": 18931.27801723968,
+      "required_pages_per_second_upper_estimate": 0.004379417990174776
     },
     {
       "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": 28292.09494380299,
-      "required_pages_per_second_upper_estimate": 0.0017228401449272594
+      "required_token_positions_per_second": 28396.917025859522,
+      "required_pages_per_second_upper_estimate": 0.0017292232597655407
     },
     {
       "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": 28292.09494380299,
-      "required_pages_per_second_upper_estimate": 0.0021780037863053034
+      "required_token_positions_per_second": 28396.917025859522,
+      "required_pages_per_second_upper_estimate": 0.00218607328034811
     },
     {
       "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": 28292.09494380299,
-      "required_pages_per_second_upper_estimate": 0.3512931316513278
+      "required_token_positions_per_second": 28396.917025859522,
+      "required_pages_per_second_upper_estimate": 0.3525946710935287
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0866105018748442,
-    "producer_p95_seconds": 0.18760650000149326,
-    "gpu_update_seconds": 0.9772341021376337,
-    "producer_batches_per_second": 11.545943948518412,
-    "gpu_batches_per_second": 1.023296258094726,
-    "producer_headroom": 11.283090167861836,
+    "producer_batch_seconds": 0.128989227343709,
+    "producer_p95_seconds": 0.22840736499711056,
+    "gpu_update_seconds": 0.9736268192361331,
+    "producer_batches_per_second": 7.752585394866865,
+    "gpu_batches_per_second": 1.0270875660394791,
+    "producer_headroom": 7.548125058860728,
     "sustainable": true,
-    "buffer_seconds": 60.58851433253329,
+    "buffer_seconds": 60.36486279264025,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339287.75714497356,
-    "upload_bytes_per_second": 5799956.637808847,
+    "generation_bytes_per_second": 382018.72702771594,
+    "upload_bytes_per_second": 5804674.648603864,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.80325446855454,
+    "estimated_full_shards_per_hour": 81.97232587932213,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 6175962,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1872953079982835,
+      "bytes": 7019215,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -77,3 +77,4 @@
 {"step": 92700, "loss": 0.5997125413268805, "lr": 0.0003, "grad_norm": 0.1257633616212327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13565981359, "tokens_per_second": 147937.23107947153, "data_wait_s": 1.3378999938140623e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 92800, "loss": 0.6019905339926481, "lr": 0.0003, "grad_norm": 0.12667470160401403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13580623119, "tokens_per_second": 149293.2970835035, "data_wait_s": 1.1711999832186848e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 92900, "loss": 0.5813432447612286, "lr": 0.0003, "grad_norm": 0.12336659881267349, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13595227169, "tokens_per_second": 149047.19241555125, "data_wait_s": 1.6789999790489674e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 93000, "loss": 0.5979013331234455, "lr": 0.0003, "grad_norm": 0.1368685890683131, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13609897800, "tokens_per_second": 149419.68909826386, "data_wait_s": 1.0979001672239974e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:80b75f4fb0b59177e34df0749f527f4a43b5863b925f325f2eed40b83b2cac5d
-size 168270
+oid sha256:b458f627e845a1655493f3f63d61e16d142f732d358f846f70e4a8d1bfd96e65
+size 170388