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": 76587.00229088067,
-      "required_pages_per_second_upper_estimate": 0.005298466524811593
+      "required_token_positions_per_second": 76302.59609063012,
+      "required_pages_per_second_upper_estimate": 0.005278790643964958
     },
     {
       "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": 9573.375286360084,
-      "required_pages_per_second_upper_estimate": 0.004008172317589917
+      "required_token_positions_per_second": 9537.824511328765,
+      "required_pages_per_second_upper_estimate": 0.003993287950468906
     },
     {
       "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": 19146.750572720168,
-      "required_pages_per_second_upper_estimate": 0.003987763210529683
+      "required_token_positions_per_second": 19075.64902265753,
+      "required_pages_per_second_upper_estimate": 0.0039729546327256044
     },
     {
       "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": 28720.12585908025,
-      "required_pages_per_second_upper_estimate": 0.0017217276921588506
+      "required_token_positions_per_second": 28613.473533986293,
+      "required_pages_per_second_upper_estimate": 0.001715334048118139
     },
     {
       "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": 28720.12585908025,
-      "required_pages_per_second_upper_estimate": 0.0022328634590604297
+      "required_token_positions_per_second": 28613.473533986293,
+      "required_pages_per_second_upper_estimate": 0.002224571709898375
     },
     {
       "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": 28720.12585908025,
-      "required_pages_per_second_upper_estimate": 0.36839093725170596
+      "required_token_positions_per_second": 28613.473533986293,
+      "required_pages_per_second_upper_estimate": 0.36702291573974544
     }
   ],
   "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": 0.16450707424993993,
-    "producer_p95_seconds": 0.3182404899998801,
-    "gpu_update_seconds": 0.9626698760186219,
-    "producer_batches_per_second": 6.07876594097512,
-    "gpu_batches_per_second": 1.038777700342891,
-    "producer_headroom": 5.85184485474474,
+    "producer_batch_seconds": 0.08466268243748232,
+    "producer_p95_seconds": 0.2969719940010691,
+    "gpu_update_seconds": 0.9662580800321383,
+    "producer_batches_per_second": 11.811579449286084,
+    "gpu_batches_per_second": 1.0349201943716106,
+    "producer_headroom": 11.413034080814233,
     "sustainable": true,
-    "buffer_seconds": 60.648202189173176,
+    "buffer_seconds": 60.87425904202471,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382036.3434112831,
-    "upload_bytes_per_second": 4912770.5716533605,
+    "generation_bytes_per_second": 382556.1364857674,
+    "upload_bytes_per_second": 4942842.265732748,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.97610594514722,
+    "estimated_full_shards_per_hour": 82.08764143876806,
     "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": 1.1331273229989165,
-      "bytes": 4688984,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.3685046969985706,
       "bytes": 7267431,
@@ -562,6 +556,12 @@
       "bytes": 1128299,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.9414047560021572,
+      "bytes": 11291288,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -306,3 +306,4 @@
 {"step": 955600, "loss": 0.5585286021232605, "lr": 3e-05, "grad_norm": 0.3537998040043666, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139873419564, "tokens_per_second": 153427.827137965, "data_wait_s": 9.121999028138816e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 955700, "loss": 0.5493685603141785, "lr": 3e-05, "grad_norm": 0.3441077539652908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139888187009, "tokens_per_second": 153591.7385881006, "data_wait_s": 8.060997060965747e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 955800, "loss": 0.5293927192687988, "lr": 3e-05, "grad_norm": 0.34714447391276426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139902951883, "tokens_per_second": 153932.88179856248, "data_wait_s": 1.3213997590355575e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 955900, "loss": 0.5105063319206238, "lr": 3e-05, "grad_norm": 0.3532509704132093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139917613706, "tokens_per_second": 152705.56575056832, "data_wait_s": 1.0668001777958125e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79ff296762987bfceced35ff9c09fd87e5ae21699a4b17e5c52f5099dfe36ebd
-size 657207
+oid sha256:d454d44ca3b1b19b2ac60aebbd61fc507b3759e37619ac502d791ed42d5dcf53
+size 659325