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": 75163.06276019925,
-      "required_pages_per_second_upper_estimate": 0.006514943781013131
+      "required_token_positions_per_second": 75284.34697209937,
+      "required_pages_per_second_upper_estimate": 0.006525456389108611
     },
     {
       "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": 9395.382845024906,
-      "required_pages_per_second_upper_estimate": 0.00374795869037215
+      "required_token_positions_per_second": 9410.543371512422,
+      "required_pages_per_second_upper_estimate": 0.003754006451058091
     },
     {
       "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": 18790.76569004981,
-      "required_pages_per_second_upper_estimate": 0.004346912936211898
+      "required_token_positions_per_second": 18821.086743024844,
+      "required_pages_per_second_upper_estimate": 0.0043539271781853645
     },
     {
       "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": 28186.148535074717,
-      "required_pages_per_second_upper_estimate": 0.0017163885644935553
+      "required_token_positions_per_second": 28231.630114537267,
+      "required_pages_per_second_upper_estimate": 0.0017191581540594188
     },
     {
       "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": 28186.148535074717,
-      "required_pages_per_second_upper_estimate": 0.0021698477384829713
+      "required_token_positions_per_second": 28231.630114537267,
+      "required_pages_per_second_upper_estimate": 0.0021733490363710683
     },
     {
       "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": 28186.148535074717,
-      "required_pages_per_second_upper_estimate": 0.34997763183474323
+      "required_token_positions_per_second": 28231.630114537267,
+      "required_pages_per_second_upper_estimate": 0.3505423608346135
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.07373003621092167,
-    "producer_p95_seconds": 0.13981506799996168,
-    "gpu_update_seconds": 0.9809073405539942,
-    "producer_batches_per_second": 13.562993474454167,
-    "gpu_batches_per_second": 1.0194642843994037,
-    "producer_headroom": 13.304039858978014,
+    "producer_batch_seconds": 0.11928684445312143,
+    "producer_p95_seconds": 0.14085754300003828,
+    "gpu_update_seconds": 0.9793270841191442,
+    "producer_batches_per_second": 8.38315410709846,
+    "gpu_batches_per_second": 1.0211093068047332,
+    "producer_headroom": 8.209849867426161,
     "sustainable": true,
-    "buffer_seconds": 60.81625511434764,
+    "buffer_seconds": 60.718279215386936,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 151,
+    "upload_pages_per_commit": 150,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 696176.8786126114,
-    "upload_bytes_per_second": 5129351.601751318,
+    "generation_bytes_per_second": 684925.3634218505,
+    "upload_bytes_per_second": 4934365.172494287,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.6916759909809,
+    "estimated_full_shards_per_hour": 73.4845193719465,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 5982796,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.1530046430000311,
+      "bytes": 2613544,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -11,3 +11,4 @@
 {"step": 1100, "loss": 1.0527566373348236, "lr": 0.0003, "grad_norm": 0.35079080124550893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 161575502, "tokens_per_second": 148849.57722070834, "data_wait_s": 1.6026000139390817e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 1200, "loss": 1.0346743538975716, "lr": 0.0003, "grad_norm": 0.39644772406410694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 176121675, "tokens_per_second": 148757.9937526799, "data_wait_s": 1.2732000186588266e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 1300, "loss": 0.9949412643909454, "lr": 0.0003, "grad_norm": 0.24531524335142565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 190751855, "tokens_per_second": 149002.07766916553, "data_wait_s": 1.797400000214111e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 1400, "loss": 1.0136450193822384, "lr": 0.0003, "grad_norm": 0.2711707182025311, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 205319076, "tokens_per_second": 148911.63544179744, "data_wait_s": 1.1697999980242457e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fea1836cc63b2e4c310a032607c4d7e795fa588ca5c55db1aac6122603ece7e5
-size 27205
+oid sha256:a07b30e88840327e0dee8ba7c16a566f7fa7fd8a1c620cfe797c896e64e2f342
+size 29288