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": 74086.6283620379,
-      "required_pages_per_second_upper_estimate": 0.006421641175578546
+      "required_token_positions_per_second": 74759.73872587517,
+      "required_pages_per_second_upper_estimate": 0.00647998467593334
     },
     {
       "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": 9260.828545254737,
-      "required_pages_per_second_upper_estimate": 0.0036942829684277713
+      "required_token_positions_per_second": 9344.967340734396,
+      "required_pages_per_second_upper_estimate": 0.0037278471919317043
     },
     {
       "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": 18521.657090509474,
-      "required_pages_per_second_upper_estimate": 0.004284659397857833
+      "required_token_positions_per_second": 18689.934681468792,
+      "required_pages_per_second_upper_estimate": 0.004323587456941821
     },
     {
       "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": 27782.485635764213,
-      "required_pages_per_second_upper_estimate": 0.0016918076117811017
+      "required_token_positions_per_second": 28034.902022203187,
+      "required_pages_per_second_upper_estimate": 0.0017071784453888048
     },
     {
       "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": 27782.485635764213,
-      "required_pages_per_second_upper_estimate": 0.002138772651083626
+      "required_token_positions_per_second": 28034.902022203187,
+      "required_pages_per_second_upper_estimate": 0.0021582043632449766
     },
     {
       "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": 27782.485635764213,
-      "required_pages_per_second_upper_estimate": 0.34496548959812523
+      "required_token_positions_per_second": 28034.902022203187,
+      "required_pages_per_second_upper_estimate": 0.3480996563343952
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.3908052850779029,
-    "producer_p95_seconds": 0.2818436150009802,
-    "gpu_update_seconds": 0.9951593375219426,
-    "producer_batches_per_second": 2.5588190287668717,
-    "gpu_batches_per_second": 1.004864208469481,
-    "producer_headroom": 2.5464326495061806,
+    "producer_batch_seconds": 0.3393074827889393,
+    "producer_p95_seconds": 0.2730642649985384,
+    "gpu_update_seconds": 0.9861992732524348,
+    "producer_batches_per_second": 2.947179330619224,
+    "gpu_batches_per_second": 1.0139938520762148,
+    "producer_headroom": 2.9065061140012762,
     "sustainable": true,
-    "buffer_seconds": 60.7047195888385,
+    "buffer_seconds": 60.15815566839853,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372739.0945486964,
+    "generation_bytes_per_second": 339331.70434015983,
     "upload_bytes_per_second": 6243179.083831291,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.98113276811284,
+    "estimated_full_shards_per_hour": 72.81268451360317,
     "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": [

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

@@ -169,3 +169,4 @@
 {"step": 56900, "loss": 0.6293017156422138, "lr": 0.0003, "grad_norm": 0.1302298009260594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8328199059, "tokens_per_second": 149435.04813178402, "data_wait_s": 1.0220999683951959e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 57000, "loss": 0.6150839980691671, "lr": 0.0003, "grad_norm": 0.13519484046330424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8342772660, "tokens_per_second": 148430.38882894648, "data_wait_s": 1.074399915523827e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 57100, "loss": 0.6742852330207825, "lr": 0.0003, "grad_norm": 0.15001693965512816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8357399409, "tokens_per_second": 149993.12242864975, "data_wait_s": 1.1885000276379287e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 57200, "loss": 0.6306101009249687, "lr": 0.0003, "grad_norm": 0.13996838550861912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8371995313, "tokens_per_second": 148911.11273361236, "data_wait_s": 1.0081999789690599e-05, "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 sha256:6a0056f0a95fd814b86ce0222cb1eb72b3b7e4f2ceb8f46299b4fc1b1093f1ed
-size 364692
+oid sha256:d2b5bef4816812b34b80d0ef872068806d52d39951da375a3626e278f153b88e
+size 366810