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": 76558.76322738803,
-      "required_pages_per_second_upper_estimate": 0.005296512880875518
+      "required_token_positions_per_second": 77039.38278939603,
+      "required_pages_per_second_upper_estimate": 0.005329763257366259
     },
     {
       "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": 9569.845403423504,
-      "required_pages_per_second_upper_estimate": 0.004006694429316709
+      "required_token_positions_per_second": 9629.922848674503,
+      "required_pages_per_second_upper_estimate": 0.004031847601083585
     },
     {
       "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": 19139.690806847007,
-      "required_pages_per_second_upper_estimate": 0.00398629284747685
+      "required_token_positions_per_second": 19259.845697349007,
+      "required_pages_per_second_upper_estimate": 0.004011317942470868
     },
     {
       "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": 28709.53621027051,
-      "required_pages_per_second_upper_estimate": 0.0017210928588821642
+      "required_token_positions_per_second": 28889.768546023508,
+      "required_pages_per_second_upper_estimate": 0.0017318975122117143
     },
     {
       "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": 28709.53621027051,
-      "required_pages_per_second_upper_estimate": 0.002232040160444415
+      "required_token_positions_per_second": 28889.768546023508,
+      "required_pages_per_second_upper_estimate": 0.002246052431790947
     },
     {
       "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": 28709.53621027051,
-      "required_pages_per_second_upper_estimate": 0.3682551046070537
+      "required_token_positions_per_second": 28889.768546023508,
+      "required_pages_per_second_upper_estimate": 0.37056693149168823
     }
   ],
   "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": 0.3121992248749166,
-    "producer_p95_seconds": 0.3109045159944799,
-    "gpu_update_seconds": 0.963024961375351,
-    "producer_batches_per_second": 3.2030829045160267,
-    "gpu_batches_per_second": 1.0383946835311961,
-    "producer_headroom": 3.0846487904035937,
+    "producer_batch_seconds": 0.4536370959920646,
+    "producer_p95_seconds": 0.2912266159983119,
+    "gpu_update_seconds": 0.9570170129938812,
+    "producer_batches_per_second": 2.204405258818368,
+    "gpu_batches_per_second": 1.0449135035454105,
+    "producer_headroom": 2.109653336222358,
     "sustainable": true,
-    "buffer_seconds": 60.67057256664711,
+    "buffer_seconds": 60.29207181861452,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 410864.09113209136,
+    "generation_bytes_per_second": 363779.76039098576,
     "upload_bytes_per_second": 4912981.942445658,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.1618695304113,
+    "estimated_full_shards_per_hour": 78.05866822049312,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 794700, "loss": 0.5018172860145569, "lr": 3e-05, "grad_norm": 0.27744146885200277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116321529952, "tokens_per_second": 151068.95150713107, "data_wait_s": 8.745002560317516e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 794800, "loss": 0.506133496761322, "lr": 3e-05, "grad_norm": 0.27428337174528683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116336077825, "tokens_per_second": 151475.28008538822, "data_wait_s": 9.927003702614456e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 794900, "loss": 0.5389185547828674, "lr": 3e-05, "grad_norm": 0.2768717896283943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116350658479, "tokens_per_second": 151620.15939716657, "data_wait_s": 1.3008997484575957e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 795000, "loss": 0.4829784333705902, "lr": 3e-05, "grad_norm": 0.26054928311181613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116365386659, "tokens_per_second": 152847.1465283554, "data_wait_s": 1.2873999367002398e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d792e52c59331d7d81db2be8c5d5a85be117891d55ac0b909ed641742cbdd7af
-size 744045
+oid sha256:b893c3796cc5bd604ff395301cce586336baa174db52061638414a614468981c
+size 746163