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": 76596.72193814552,
-      "required_pages_per_second_upper_estimate": 0.005299138952562065
+      "required_token_positions_per_second": 77110.63162591206,
+      "required_pages_per_second_upper_estimate": 0.0053346924171965154
     },
     {
       "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": 9574.59024226819,
-      "required_pages_per_second_upper_estimate": 0.004008680994257477
+      "required_token_positions_per_second": 9638.828953239008,
+      "required_pages_per_second_upper_estimate": 0.004035576401084131
     },
     {
       "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": 19149.18048453638,
-      "required_pages_per_second_upper_estimate": 0.003988269297079916
+      "required_token_positions_per_second": 19277.657906478016,
+      "required_pages_per_second_upper_estimate": 0.004015027755892898
     },
     {
       "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": 28723.770726804567,
-      "required_pages_per_second_upper_estimate": 0.0017219461964135326
+      "required_token_positions_per_second": 28916.486859717028,
+      "required_pages_per_second_upper_estimate": 0.001733499234321138
     },
     {
       "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": 28723.770726804567,
-      "required_pages_per_second_upper_estimate": 0.0022331468314939163
+      "required_token_positions_per_second": 28916.486859717028,
+      "required_pages_per_second_upper_estimate": 0.002248129663159176
     },
     {
       "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": 28723.770726804567,
-      "required_pages_per_second_upper_estimate": 0.36843768970131946
+      "required_token_positions_per_second": 28916.486859717028,
+      "required_pages_per_second_upper_estimate": 0.3709096453318586
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5243184016484577,
-    "producer_p95_seconds": 0.2794876150001073,
-    "gpu_update_seconds": 0.962547719203152,
-    "producer_batches_per_second": 1.9072380386726822,
-    "gpu_batches_per_second": 1.0389095314961143,
-    "producer_headroom": 1.8358076241018833,
+    "producer_batch_seconds": 0.10149411802359509,
+    "producer_p95_seconds": 0.27157277399965096,
+    "gpu_update_seconds": 0.9561327464891964,
+    "producer_batches_per_second": 9.852787722806976,
+    "gpu_batches_per_second": 1.0458798777386078,
+    "producer_headroom": 9.42057298598247,
     "sustainable": true,
-    "buffer_seconds": 60.640506309798575,
+    "buffer_seconds": 60.236363028819376,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 220447.39200162108,
+    "generation_bytes_per_second": 248659.81048031786,
     "upload_bytes_per_second": 4699773.933401427,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 47.30287857090448,
+    "estimated_full_shards_per_hour": 53.356606824943086,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -299,3 +299,4 @@
 {"step": 914900, "loss": 0.5532621741294861, "lr": 3e-05, "grad_norm": 0.33788388176720446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133914855868, "tokens_per_second": 151628.29680967875, "data_wait_s": 9.654999303165823e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 915000, "loss": 0.5263814926147461, "lr": 3e-05, "grad_norm": 0.31579094226674975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133929383665, "tokens_per_second": 136591.20693712513, "data_wait_s": 1.737300044624135e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 915100, "loss": 0.5375716686248779, "lr": 3e-05, "grad_norm": 0.33293834218421364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133943923327, "tokens_per_second": 144171.80356261818, "data_wait_s": 1.591299951542169e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 915200, "loss": 0.5317983031272888, "lr": 3e-05, "grad_norm": 0.31753135484635203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133958596336, "tokens_per_second": 153445.90836459125, "data_wait_s": 1.0946998372673988e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f555a59e66b6175369754c978488958fd6515a9af6ad4c56773e4420fc35e2b2
-size 642381
+oid sha256:ed49593f12dc6907c8d183bd3c1b81d2959698dac0f738bd63eeac42f668a73d
+size 644499