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": 76863.92159152548,
-      "required_pages_per_second_upper_estimate": 0.005317624444571504
+      "required_token_positions_per_second": 76558.76322738803,
+      "required_pages_per_second_upper_estimate": 0.005296512880875518
     },
     {
       "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": 9607.990198940684,
-      "required_pages_per_second_upper_estimate": 0.004022664858645843
+      "required_token_positions_per_second": 9569.845403423504,
+      "required_pages_per_second_upper_estimate": 0.004006694429316709
     },
     {
       "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": 19215.98039788137,
-      "required_pages_per_second_upper_estimate": 0.004002181957397498
+      "required_token_positions_per_second": 19139.690806847007,
+      "required_pages_per_second_upper_estimate": 0.00398629284747685
     },
     {
       "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": 28823.970596822055,
-      "required_pages_per_second_upper_estimate": 0.0017279530256247384
+      "required_token_positions_per_second": 28709.53621027051,
+      "required_pages_per_second_upper_estimate": 0.0017210928588821642
     },
     {
       "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": 28823.970596822055,
-      "required_pages_per_second_upper_estimate": 0.002240936930655126
+      "required_token_positions_per_second": 28709.53621027051,
+      "required_pages_per_second_upper_estimate": 0.002232040160444415
     },
     {
       "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": 28823.970596822055,
-      "required_pages_per_second_upper_estimate": 0.3697229460476656
+      "required_token_positions_per_second": 28709.53621027051,
+      "required_pages_per_second_upper_estimate": 0.3682551046070537
     }
   ],
   "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": 0.0862196733284577,
-    "producer_p95_seconds": 0.3022764260022086,
-    "gpu_update_seconds": 0.959201644587033,
-    "producer_batches_per_second": 11.598281011695038,
-    "gpu_batches_per_second": 1.0425336587392235,
-    "producer_headroom": 11.125090220800438,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.42970360898308,
+    "buffer_seconds": 60.67057256664711,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 433150.11736258643,
-    "upload_bytes_per_second": 4854012.947128759,
+    "generation_bytes_per_second": 410864.09113209136,
+    "upload_bytes_per_second": 4912981.942445658,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.94393196733661,
+    "estimated_full_shards_per_hour": 88.1618695304113,
     "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": 1.3943405380014156,
-      "bytes": 6107726,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.2131986140011577,
       "bytes": 10368948,
@@ -562,6 +556,12 @@
       "bytes": 6307452,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.4970643039996503,
+      "bytes": 11855130,
+      "pages": 30,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -346,3 +346,4 @@
 {"step": 794600, "loss": 0.508455753326416, "lr": 3e-05, "grad_norm": 0.2578542515966335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116307039693, "tokens_per_second": 151368.67870592087, "data_wait_s": 9.748000593390316e-06, "gpu_peak_gb": 89.37899398803711}
 {"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}

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:f239dd02f1c77877767339ef9bb5beafd024875953a5195df8f0ec3e848bb774
-size 741927
+oid sha256:d792e52c59331d7d81db2be8c5d5a85be117891d55ac0b909ed641742cbdd7af
+size 744045