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": 75720.5920557672,
-      "required_pages_per_second_upper_estimate": 0.006563269007307882
+      "required_token_positions_per_second": 75294.19321551842,
+      "required_pages_per_second_upper_estimate": 0.006526309836533097
     },
     {
       "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": 9465.0740069709,
-      "required_pages_per_second_upper_estimate": 0.0037757595368481333
+      "required_token_positions_per_second": 9411.774151939802,
+      "required_pages_per_second_upper_estimate": 0.00375449742777238
     },
     {
       "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": 18930.1480139418,
-      "required_pages_per_second_upper_estimate": 0.004379156583799191
+      "required_token_positions_per_second": 18823.548303879605,
+      "required_pages_per_second_upper_estimate": 0.004354496616967123
     },
     {
       "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": 28395.222020912704,
-      "required_pages_per_second_upper_estimate": 0.0017291200428572833
+      "required_token_positions_per_second": 28235.32245581941,
+      "required_pages_per_second_upper_estimate": 0.0017193829982712838
     },
     {
       "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": 28395.222020912704,
-      "required_pages_per_second_upper_estimate": 0.00218594279417523
+      "required_token_positions_per_second": 28235.32245581941,
+      "required_pages_per_second_upper_estimate": 0.0021736332830240225
     },
     {
       "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": 28395.222020912704,
-      "required_pages_per_second_upper_estimate": 0.3525736248049059
+      "required_token_positions_per_second": 28235.32245581941,
+      "required_pages_per_second_upper_estimate": 0.3505882073558664
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.11163377882030545,
-    "producer_p95_seconds": 0.370146418998047,
-    "gpu_update_seconds": 0.9736849382490341,
-    "producer_batches_per_second": 8.957862132479445,
-    "gpu_batches_per_second": 1.0270262594369468,
-    "producer_headroom": 8.722135437306608,
+    "producer_batch_seconds": 0.11289917382015346,
+    "producer_p95_seconds": 0.3555639389996941,
+    "gpu_update_seconds": 0.9791990172331693,
+    "producer_batches_per_second": 8.85746074274187,
+    "gpu_batches_per_second": 1.0212428550281905,
+    "producer_headroom": 8.673216854474216,
     "sustainable": true,
-    "buffer_seconds": 60.36846617144012,
+    "buffer_seconds": 60.71033906845649,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 282226.6294643996,
-    "upload_bytes_per_second": 4227074.510656481,
+    "generation_bytes_per_second": 312717.6648123108,
+    "upload_bytes_per_second": 4152064.367452919,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.55926478337279,
+    "estimated_full_shards_per_hour": 67.10193117405885,
     "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": 1.1820991080021486,
-      "bytes": 11679771,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.2690833449996717,
       "bytes": 2379532,
@@ -562,6 +556,12 @@
       "bytes": 5908271,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8663640420018055,
+      "bytes": 2740775,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -297,3 +297,4 @@
 {"step": 344700, "loss": 0.5710683539509773, "lr": 0.0003, "grad_norm": 0.17510563112438046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50454692257, "tokens_per_second": 148467.21241022888, "data_wait_s": 1.3614997442346066e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 344800, "loss": 0.5287780500948429, "lr": 0.0003, "grad_norm": 0.1372135377320774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50469335301, "tokens_per_second": 148976.86100213657, "data_wait_s": 1.4817000192124397e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 344900, "loss": 0.5837528426200151, "lr": 0.0003, "grad_norm": 0.18961747227709821, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50483965852, "tokens_per_second": 148048.0188297243, "data_wait_s": 1.8357000953983516e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 345000, "loss": 0.5730580352246761, "lr": 0.0003, "grad_norm": 0.16349241880176352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50498610481, "tokens_per_second": 149348.46023168947, "data_wait_s": 1.3778000720776618e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02fea8058267dcce4cb91b93767e56ea5c8e7786307e3962a9dc1b8b5cdb5fb9
-size 637362
+oid sha256:4c9a74b9529482bc0a3a09d8e92de3d60c72e7d376131af0f78cdf433505faee
+size 639480