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": 74693.74685574796,
-      "required_pages_per_second_upper_estimate": 0.00647426466788558
+      "required_token_positions_per_second": 74443.06439227307,
+      "required_pages_per_second_upper_estimate": 0.0064525361473004705
     },
     {
       "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": 9336.718356968495,
-      "required_pages_per_second_upper_estimate": 0.003724556548974188
+      "required_token_positions_per_second": 9305.383049034133,
+      "required_pages_per_second_upper_estimate": 0.0037120564261345673
     },
     {
       "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": 18673.43671393699,
-      "required_pages_per_second_upper_estimate": 0.00431977094250765
+      "required_token_positions_per_second": 18610.766098068267,
+      "required_pages_per_second_upper_estimate": 0.00430527319849159
     },
     {
       "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": 28010.155070905486,
-      "required_pages_per_second_upper_estimate": 0.0017056714858919941
+      "required_token_positions_per_second": 27916.149147102402,
+      "required_pages_per_second_upper_estimate": 0.001699947018343351
     },
     {
       "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": 28010.155070905486,
-      "required_pages_per_second_upper_estimate": 0.002156299274430134
+      "required_token_positions_per_second": 27916.149147102402,
+      "required_pages_per_second_upper_estimate": 0.0021490624381907204
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28010.155070905486,
-      "required_pages_per_second_upper_estimate": 0.3477923820219959
+      "required_token_positions_per_second": 27916.149147102402,
+      "required_pages_per_second_upper_estimate": 0.3466251430659498
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.13179184497650454,
-    "producer_p95_seconds": 0.3187793329998385,
-    "gpu_update_seconds": 0.9870705795812726,
-    "producer_batches_per_second": 7.587722898774785,
-    "gpu_batches_per_second": 1.0130987800530051,
-    "producer_headroom": 7.489618039395721,
+    "producer_batch_seconds": 0.10182360034400517,
+    "producer_p95_seconds": 0.2887334290026047,
+    "gpu_update_seconds": 0.9903944793499488,
+    "producer_batches_per_second": 9.820905925753534,
+    "gpu_batches_per_second": 1.0096986815358218,
+    "producer_headroom": 9.726571011081498,
     "sustainable": true,
-    "buffer_seconds": 60.21130535445763,
+    "buffer_seconds": 60.41406324034688,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 447743.45097252086,
-    "upload_bytes_per_second": 2706899.619465454,
+    "generation_bytes_per_second": 444128.5697524922,
+    "upload_bytes_per_second": 2715922.171041771,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.07532164460868,
+    "estimated_full_shards_per_hour": 95.29965228491854,
     "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": 1.7056977829997777,
-      "bytes": 6292570,
-      "pages": 16,
-      "shards": 2
-    },
     {
       "seconds": 10.767963721002161,
       "bytes": 5800923,
@@ -562,6 +556,12 @@
       "bytes": 2767942,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0177543880017765,
+      "bytes": 5824791,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -224,3 +224,4 @@
 {"step": 302600, "loss": 0.5477177314460278, "lr": 0.0003, "grad_norm": 0.1288385868721625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44292791524, "tokens_per_second": 148154.17344228033, "data_wait_s": 1.772099858499132e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 302700, "loss": 0.6579298116266727, "lr": 0.0003, "grad_norm": 0.1791506841409226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44307495754, "tokens_per_second": 150504.77405587974, "data_wait_s": 1.802999759092927e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 302800, "loss": 0.59256577398628, "lr": 0.0003, "grad_norm": 0.15586981353577598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44322239027, "tokens_per_second": 150793.86006886125, "data_wait_s": 1.1732001439668238e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 302900, "loss": 0.6034968458116055, "lr": 0.0003, "grad_norm": 0.1457596251327611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44336969250, "tokens_per_second": 150457.0899802314, "data_wait_s": 1.020499985315837e-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:a4aee8cd64e05793080248b669963e34c65d96f39fdb8a1a0037fafeaf6d106c
-size 481965
+oid sha256:a4dbf513c0ad6ffb7cb5ec2432bc02b795bc4c84847fb4444f0f5639afb29985
+size 484083