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": 77375.13936191444,
-      "required_pages_per_second_upper_estimate": 0.005352991676115664
+      "required_token_positions_per_second": 77316.83304675309,
+      "required_pages_per_second_upper_estimate": 0.005348957909943512
     },
     {
       "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": 9671.892420239305,
-      "required_pages_per_second_upper_estimate": 0.004049419384273451
+      "required_token_positions_per_second": 9664.604130844136,
+      "required_pages_per_second_upper_estimate": 0.004046367929700484
     },
     {
       "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": 19343.78484047861,
-      "required_pages_per_second_upper_estimate": 0.0040288002523607
+      "required_token_positions_per_second": 19329.208261688273,
+      "required_pages_per_second_upper_estimate": 0.004025764335408907
     },
     {
       "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": 29015.677260717912,
-      "required_pages_per_second_upper_estimate": 0.0017394455474061682
+      "required_token_positions_per_second": 28993.812392532407,
+      "required_pages_per_second_upper_estimate": 0.0017381347819441677
     },
     {
       "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": 29015.677260717912,
-      "required_pages_per_second_upper_estimate": 0.0022558412805445296
+      "required_token_positions_per_second": 28993.812392532407,
+      "required_pages_per_second_upper_estimate": 0.002254141383216498
     },
     {
       "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": 29015.677260717912,
-      "required_pages_per_second_upper_estimate": 0.3721819532935431
+      "required_token_positions_per_second": 28993.812392532407,
+      "required_pages_per_second_upper_estimate": 0.3719014942411258
     }
   ],
   "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": 0.08089001985932498,
+    "producer_batch_seconds": 0.05058113979686851,
     "producer_p95_seconds": 0.17515987800015864,
-    "gpu_update_seconds": 0.9528641965366252,
-    "producer_batches_per_second": 12.362464513410801,
-    "gpu_batches_per_second": 1.0494674935155495,
-    "producer_headroom": 11.779749815783724,
+    "gpu_update_seconds": 0.9535827722718164,
+    "producer_batches_per_second": 19.77021482742289,
+    "gpu_batches_per_second": 1.0486766635030529,
+    "producer_headroom": 18.852536263543293,
     "sustainable": true,
-    "buffer_seconds": 60.03044438180739,
+    "buffer_seconds": 60.075714653124436,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333403.48656474106,
+    "generation_bytes_per_second": 389019.5003870405,
     "upload_bytes_per_second": 4116738.182318712,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.5406269808452,
+    "estimated_full_shards_per_hour": 83.47452887257015,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -77,3 +77,4 @@
 {"step": 932700, "loss": 0.5129926800727844, "lr": 3e-05, "grad_norm": 0.3344476085776876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136520966178, "tokens_per_second": 150316.800938626, "data_wait_s": 1.321299987466773e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 932800, "loss": 0.5275412797927856, "lr": 3e-05, "grad_norm": 0.35713417561118094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136535596977, "tokens_per_second": 152941.3422571803, "data_wait_s": 1.138099923991831e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 932900, "loss": 0.5345956087112427, "lr": 3e-05, "grad_norm": 0.3351441269328984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136550277302, "tokens_per_second": 153532.8995836737, "data_wait_s": 8.703999810677487e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 933000, "loss": 0.5141292810440063, "lr": 3e-05, "grad_norm": 0.346287447739729, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136564981385, "tokens_per_second": 152895.9097757887, "data_wait_s": 1.5439999515365344e-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 sha256:107a2ede5f6d7c618be671beb0d6026b97c1bac7c363294f43f172d48ac22827
-size 168270
+oid sha256:dcea60d64dbcf756f4ba0ad07276ead18b56f4a5be28471bad2b697f36c1ae80
+size 170388