Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76155.01078101793,
-      "required_pages_per_second_upper_estimate": 0.006600923321388422
+      "required_token_positions_per_second": 76073.84723048519,
+      "required_pages_per_second_upper_estimate": 0.006593888270535381
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9519.376347627242,
-      "required_pages_per_second_upper_estimate": 0.0037974215524282915
+      "required_token_positions_per_second": 9509.230903810649,
+      "required_pages_per_second_upper_estimate": 0.00379337438320195
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19038.752695254483,
-      "required_pages_per_second_upper_estimate": 0.00440428036544377
+      "required_token_positions_per_second": 19018.461807621297,
+      "required_pages_per_second_upper_estimate": 0.004399586425697258
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28558.129042881726,
-      "required_pages_per_second_upper_estimate": 0.0017390402257881077
+      "required_token_positions_per_second": 28527.692711431948,
+      "required_pages_per_second_upper_estimate": 0.001737186812889906
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28558.129042881726,
-      "required_pages_per_second_upper_estimate": 0.0021984838276819046
+      "required_token_positions_per_second": 28527.692711431948,
+      "required_pages_per_second_upper_estimate": 0.0021961407546337427
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28558.129042881726,
-      "required_pages_per_second_upper_estimate": 0.35459638480303124
+      "required_token_positions_per_second": 28527.692711431948,
+      "required_pages_per_second_upper_estimate": 0.35421846743027363
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06285995994522864,
-    "producer_p95_seconds": 0.20234570899992832,
-    "gpu_update_seconds": 0.9681306488420472,
-    "producer_batches_per_second": 15.908377938377999,
-    "gpu_batches_per_second": 1.0329184404977476,
-    "producer_headroom": 15.401388255506403,
+    "producer_batch_seconds": 0.11307417841410938,
+    "producer_p95_seconds": 0.21705139700134168,
+    "gpu_update_seconds": 0.96916355204992,
+    "producer_batches_per_second": 8.843752075188371,
+    "gpu_batches_per_second": 1.0318175893891763,
+    "producer_headroom": 8.571042174638412,
     "sustainable": true,
-    "buffer_seconds": 60.024100228206926,
+    "buffer_seconds": 60.08814022709504,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373637.31285776687,
+    "generation_bytes_per_second": 373675.14529498067,
     "upload_bytes_per_second": 5504907.379593526,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.17386950778727,
+    "estimated_full_shards_per_hour": 80.18198746811929,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -60,3 +60,4 @@
 {"step": 286200, "loss": 0.6156712435185909, "lr": 0.0003, "grad_norm": 0.15521044479639703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41891574606, "tokens_per_second": 150345.41218694282, "data_wait_s": 1.8108001313521527e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 286300, "loss": 0.5622965320944786, "lr": 0.0003, "grad_norm": 0.14726805877539556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41906283753, "tokens_per_second": 150687.9089786302, "data_wait_s": 1.4015000488143414e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 286400, "loss": 0.6382241677492857, "lr": 0.0003, "grad_norm": 0.15030694681059412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41921058959, "tokens_per_second": 151047.22960489857, "data_wait_s": 1.3300999853527173e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 286500, "loss": 0.6280013173818588, "lr": 0.0003, "grad_norm": 0.13263722039132297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41935690374, "tokens_per_second": 150288.80881129822, "data_wait_s": 1.2853999578510411e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5b10a7b24c05c1337d763a7a8e37afa807a7a4b490ab8f5de092c5a01ee2d2bf
-size 132264
+oid sha256:ddd8c8e5ac6f7c7388ada53aa5b88f67b1942cca4b30994f8847e465988007c6
+size 134382