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": 76073.84723048519,
-      "required_pages_per_second_upper_estimate": 0.006593888270535381
+      "required_token_positions_per_second": 74526.45725088581,
+      "required_pages_per_second_upper_estimate": 0.006459764428927756
     },
     {
       "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": 9509.230903810649,
-      "required_pages_per_second_upper_estimate": 0.00379337438320195
+      "required_token_positions_per_second": 9315.807156360726,
+      "required_pages_per_second_upper_estimate": 0.0037162147584014387
     },
     {
       "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": 19018.461807621297,
-      "required_pages_per_second_upper_estimate": 0.004399586425697258
+      "required_token_positions_per_second": 18631.614312721453,
+      "required_pages_per_second_upper_estimate": 0.0043100960660355575
     },
     {
       "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": 28527.692711431948,
-      "required_pages_per_second_upper_estimate": 0.001737186812889906
+      "required_token_positions_per_second": 27947.42146908218,
+      "required_pages_per_second_upper_estimate": 0.0017018513386787262
     },
     {
       "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": 28527.692711431948,
-      "required_pages_per_second_upper_estimate": 0.0021961407546337427
+      "required_token_positions_per_second": 27947.42146908218,
+      "required_pages_per_second_upper_estimate": 0.0021514698681040514
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28527.692711431948,
-      "required_pages_per_second_upper_estimate": 0.35421846743027363
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27947.42146908218,
+      "required_pages_per_second_upper_estimate": 0.3470134406432097
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "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,
+    "producer_batch_seconds": 0.48991867437501924,
+    "producer_p95_seconds": 0.2169164679999085,
+    "gpu_update_seconds": 0.9892862577889905,
+    "producer_batches_per_second": 2.0411550983960405,
+    "gpu_batches_per_second": 1.0108297695703914,
+    "producer_headroom": 2.0192866888591374,
     "sustainable": true,
-    "buffer_seconds": 60.08814022709504,
+    "buffer_seconds": 60.34646172512842,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373675.14529498067,
-    "upload_bytes_per_second": 5504907.379593526,
+    "generation_bytes_per_second": 338965.5936144524,
+    "upload_bytes_per_second": 5522070.427655375,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.18198746811929,
+    "estimated_full_shards_per_hour": 72.73412567448787,
     "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": [
@@ -484,6 +484,12 @@
       "bytes": 1188820,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1932972649992735,
+      "bytes": 7607279,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -61,3 +61,4 @@
 {"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}
+{"step": 286600, "loss": 0.5946786962449551, "lr": 0.0003, "grad_norm": 0.17412314477500343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41950326801, "tokens_per_second": 149826.42959070986, "data_wait_s": 1.4883000403642654e-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:ddd8c8e5ac6f7c7388ada53aa5b88f67b1942cca4b30994f8847e465988007c6
-size 134382
+oid sha256:c2156ddc2f0c7f31bd58e73f7201f147d13f07db91931c9b580b6cbc6dc9ab98
+size 136500