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": 75088.9614643627,
-      "required_pages_per_second_upper_estimate": 0.006508520868498037
+      "required_token_positions_per_second": 75924.62247055197,
+      "required_pages_per_second_upper_estimate": 0.006580953846551067
     },
     {
       "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": 9386.120183045337,
-      "required_pages_per_second_upper_estimate": 0.0037442636760193623
+      "required_token_positions_per_second": 9490.577808818996,
+      "required_pages_per_second_upper_estimate": 0.003785933384721157
     },
     {
       "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": 18772.240366090675,
-      "required_pages_per_second_upper_estimate": 0.004342627428548518
+      "required_token_positions_per_second": 18981.155617637993,
+      "required_pages_per_second_upper_estimate": 0.00439095629521114
     },
     {
       "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": 28158.360549136014,
-      "required_pages_per_second_upper_estimate": 0.001714696421942183
+      "required_token_positions_per_second": 28471.73342645699,
+      "required_pages_per_second_upper_estimate": 0.001733779186030609
     },
     {
       "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": 28158.360549136014,
-      "required_pages_per_second_upper_estimate": 0.0021677085424033414
+      "required_token_positions_per_second": 28471.73342645699,
+      "required_pages_per_second_upper_estimate": 0.0021918328539711554
     },
     {
       "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": 28158.360549136014,
-      "required_pages_per_second_upper_estimate": 0.34963259804979097
+      "required_token_positions_per_second": 28471.73342645699,
+      "required_pages_per_second_upper_estimate": 0.35352364039456385
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1455054930939923,
-    "producer_p95_seconds": 0.39122707700153114,
-    "gpu_update_seconds": 0.9818753457522699,
-    "producer_batches_per_second": 6.872592771147335,
-    "gpu_batches_per_second": 1.0184592212505792,
-    "producer_headroom": 6.74802940338484,
+    "producer_batch_seconds": 0.1117822228597447,
+    "producer_p95_seconds": 0.3894691659952514,
+    "gpu_update_seconds": 0.9710683780956045,
+    "producer_batches_per_second": 8.94596631214535,
+    "gpu_batches_per_second": 1.0297935990472002,
+    "producer_headroom": 8.687144997232902,
     "sustainable": true,
-    "buffer_seconds": 60.876271436640735,
+    "buffer_seconds": 60.20623944192748,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 484826.04428306036,
-    "upload_bytes_per_second": 5722691.865377483,
+    "generation_bytes_per_second": 467276.861647215,
+    "upload_bytes_per_second": 5676632.052132874,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.03238293045862,
+    "estimated_full_shards_per_hour": 100.26673686086976,
     "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": 1.2625757009991503,
-      "bytes": 11728601,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 0.9978417950005678,
       "bytes": 5753379,
@@ -562,6 +556,12 @@
       "bytes": 11962393,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.1221481789980317,
+      "bytes": 7553276,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -321,3 +321,4 @@
 {"step": 312300, "loss": 0.6069875601679087, "lr": 0.0003, "grad_norm": 0.19347847657333106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45712039954, "tokens_per_second": 150738.92517254225, "data_wait_s": 1.0047995601780713e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 312400, "loss": 0.5539985727518797, "lr": 0.0003, "grad_norm": 0.14900928106209144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45726875311, "tokens_per_second": 152208.859573901, "data_wait_s": 1.097100175684318e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 312500, "loss": 0.5605333168059587, "lr": 0.0003, "grad_norm": 0.15028700770938597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45741536649, "tokens_per_second": 150805.988811795, "data_wait_s": 9.02800093172118e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 312600, "loss": 0.5956666208803654, "lr": 0.0003, "grad_norm": 0.18772054048968298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45756026804, "tokens_per_second": 148709.25539628958, "data_wait_s": 1.050400169333443e-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 sha256:73d40d61567e8bef7423e877f3b00d0b62b3073f4be64baa9af6adb522b085d1
-size 688977
+oid sha256:f8c067cb0d6ad71382ede9b1d07b446f0c17b9d0607f6bda1084d5cbf4aeb033
+size 691095