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": 77204.39636658129,
-      "required_pages_per_second_upper_estimate": 0.005341179279520185
+      "required_token_positions_per_second": 76877.53323326915,
+      "required_pages_per_second_upper_estimate": 0.005318566129530691
     },
     {
       "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": 9650.549545822661,
-      "required_pages_per_second_upper_estimate": 0.004040483568445102
+      "required_token_positions_per_second": 9609.691654158643,
+      "required_pages_per_second_upper_estimate": 0.004023377222415177
     },
     {
       "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": 19301.099091645323,
-      "required_pages_per_second_upper_estimate": 0.004019909936577623
+      "required_token_positions_per_second": 19219.383308317287,
+      "required_pages_per_second_upper_estimate": 0.004002890693900516
     },
     {
       "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": 28951.648637467988,
-      "required_pages_per_second_upper_estimate": 0.0017356071292083822
+      "required_token_positions_per_second": 28829.07496247593,
+      "required_pages_per_second_upper_estimate": 0.0017282590245517732
     },
     {
       "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": 28951.648637467988,
-      "required_pages_per_second_upper_estimate": 0.002250863336718998
+      "required_token_positions_per_second": 28829.07496247593,
+      "required_pages_per_second_upper_estimate": 0.002241333772632983
     },
     {
       "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": 28951.648637467988,
-      "required_pages_per_second_upper_estimate": 0.3713606628630724
+      "required_token_positions_per_second": 28829.07496247593,
+      "required_pages_per_second_upper_estimate": 0.36978841936963264
     }
   ],
   "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": 0.05473698420303208,
-    "producer_p95_seconds": 0.18668323199926817,
-    "gpu_update_seconds": 0.9549715232527083,
-    "producer_batches_per_second": 18.269183342121476,
-    "gpu_batches_per_second": 1.0471516434269381,
-    "producer_headroom": 17.44654984480875,
+    "producer_batch_seconds": 0.1053770202108808,
+    "producer_p95_seconds": 0.21990923599878442,
+    "gpu_update_seconds": 0.959031812015709,
+    "producer_batches_per_second": 9.489735029504507,
+    "gpu_batches_per_second": 1.0427182784460334,
+    "producer_headroom": 9.100957780894655,
     "sustainable": true,
-    "buffer_seconds": 60.16320596492062,
+    "buffer_seconds": 60.419004156989665,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269091.872746571,
-    "upload_bytes_per_second": 4757799.075449274,
+    "generation_bytes_per_second": 270113.0246102854,
+    "upload_bytes_per_second": 4812101.186676659,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.74085175321434,
+    "estimated_full_shards_per_hour": 57.95996717196867,
     "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": 0.8907806349998282,
-      "bytes": 954545,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 1.3382855260006181,
       "bytes": 6196201,
@@ -562,6 +556,12 @@
       "bytes": 6111968,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2683859379994828,
+      "bytes": 7618531,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -114,3 +114,4 @@
 {"step": 853900, "loss": 0.5904368758201599, "lr": 3e-05, "grad_norm": 0.3303986266848577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124987376595, "tokens_per_second": 154703.17878011853, "data_wait_s": 1.7057000150089152e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 854000, "loss": 0.5267199277877808, "lr": 3e-05, "grad_norm": 0.2879479074167405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125001931071, "tokens_per_second": 151603.668125454, "data_wait_s": 1.1304000508971512e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 854100, "loss": 0.5018497705459595, "lr": 3e-05, "grad_norm": 0.3152239745849124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125016457430, "tokens_per_second": 151628.98132593484, "data_wait_s": 9.123999916482717e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 854200, "loss": 0.5001006722450256, "lr": 3e-05, "grad_norm": 0.30517504769892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125030910501, "tokens_per_second": 150413.6043681169, "data_wait_s": 1.2000000424450263e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:640567e88145407892e59a610cd5f52a46d81bf005e7073e9c21d34a37c72da4
-size 247419
+oid sha256:a739a48b4c616177d1f27057c42900e4a261d3c57b7004c8974066bbb5270e84
+size 249537