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": 75864.97720129268,
-      "required_pages_per_second_upper_estimate": 0.006575783945781224
+      "required_token_positions_per_second": 75703.16726219824,
+      "required_pages_per_second_upper_estimate": 0.006561758670364059
     },
     {
       "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": 9483.122150161586,
-      "required_pages_per_second_upper_estimate": 0.0037829592110106853
+      "required_token_positions_per_second": 9462.89590777478,
+      "required_pages_per_second_upper_estimate": 0.003774890660513316
     },
     {
       "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": 18966.24430032317,
-      "required_pages_per_second_upper_estimate": 0.004387506824380578
+      "required_token_positions_per_second": 18925.79181554956,
+      "required_pages_per_second_upper_estimate": 0.004378148853967619
     },
     {
       "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": 28449.366450484755,
-      "required_pages_per_second_upper_estimate": 0.0017324171545443534
+      "required_token_positions_per_second": 28388.68772332434,
+      "required_pages_per_second_upper_estimate": 0.0017287221384169617
     },
     {
       "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": 28449.366450484755,
-      "required_pages_per_second_upper_estimate": 0.0021901109822450603
+      "required_token_positions_per_second": 28388.68772332434,
+      "required_pages_per_second_upper_estimate": 0.0021854397658587843
     },
     {
       "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": 28449.366450484755,
-      "required_pages_per_second_upper_estimate": 0.35324591741044187
+      "required_token_positions_per_second": 28388.68772332434,
+      "required_pages_per_second_upper_estimate": 0.35249249069774563
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.08873336534378495,
-    "producer_p95_seconds": 0.21821061999798985,
-    "gpu_update_seconds": 0.971831834924004,
-    "producer_batches_per_second": 11.269717947985411,
-    "gpu_batches_per_second": 1.0289846083074636,
-    "producer_headroom": 10.952270672466645,
+    "producer_batch_seconds": 0.11651281558596338,
+    "producer_p95_seconds": 0.29949219899935997,
+    "gpu_update_seconds": 0.9739090538265428,
+    "producer_batches_per_second": 8.582746841802978,
+    "gpu_batches_per_second": 1.0267899205484787,
+    "producer_headroom": 8.358814855933087,
     "sustainable": true,
-    "buffer_seconds": 60.25357376528825,
+    "buffer_seconds": 60.382361337245655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 528544.2179385388,
-    "upload_bytes_per_second": 5757631.034889706,
+    "generation_bytes_per_second": 546147.6821233936,
+    "upload_bytes_per_second": 5694831.882575059,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.4132852899277,
+    "estimated_full_shards_per_hour": 117.1905789163242,
     "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": 1.1222496679984033,
-      "bytes": 5615350,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.8820894919990678,
       "bytes": 2581402,
@@ -562,6 +556,12 @@
       "bytes": 6228835,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0997196959979192,
+      "bytes": 878203,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -177,3 +177,4 @@
 {"step": 372700, "loss": 0.44184691086411476, "lr": 0.0003, "grad_norm": 0.1729692581375387, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54552137245, "tokens_per_second": 148941.18205493974, "data_wait_s": 9.749001037562266e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 372800, "loss": 0.6418619491159916, "lr": 0.0003, "grad_norm": 0.14596451638189695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54566934031, "tokens_per_second": 151031.66069808954, "data_wait_s": 8.03800139692612e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 372900, "loss": 0.5934477001428604, "lr": 0.0003, "grad_norm": 0.16541215916889493, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54581644488, "tokens_per_second": 149593.7092877425, "data_wait_s": 1.3289998605614528e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 373000, "loss": 0.5714317765086889, "lr": 0.0003, "grad_norm": 0.14223743901543276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54596359369, "tokens_per_second": 150206.74680427508, "data_wait_s": 1.1932999768760055e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e979b82a067aabcd51f27f25a115eb8271b969f874108d78efb58d1fa443848
-size 381636
+oid sha256:3694300ed5312f69c50d36e10fefe9e78fbb0a31e22856127746cf0886c57cd0
+size 383754