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": 75758.29940357714,
-      "required_pages_per_second_upper_estimate": 0.006566537384647648
+      "required_token_positions_per_second": 75864.14331186358,
+      "required_pages_per_second_upper_estimate": 0.0065757116663589825
     },
     {
       "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": 9469.787425447143,
-      "required_pages_per_second_upper_estimate": 0.0037776397899501927
+      "required_token_positions_per_second": 9483.017913982947,
+      "required_pages_per_second_upper_estimate": 0.0037829176296405565
     },
     {
       "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": 18939.574850894285,
-      "required_pages_per_second_upper_estimate": 0.004381337316621484
+      "required_token_positions_per_second": 18966.035827965894,
+      "required_pages_per_second_upper_estimate": 0.00438745859796971
     },
     {
       "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": 28409.362276341428,
-      "required_pages_per_second_upper_estimate": 0.0017299811102247057
+      "required_token_positions_per_second": 28449.053741948843,
+      "required_pages_per_second_upper_estimate": 0.0017323981122351713
     },
     {
       "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": 28409.362276341428,
-      "required_pages_per_second_upper_estimate": 0.0021870313501808655
+      "required_token_positions_per_second": 28449.053741948843,
+      "required_pages_per_second_upper_estimate": 0.0021900869090763333
     },
     {
       "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": 28409.362276341428,
-      "required_pages_per_second_upper_estimate": 0.35274919945294
+      "required_token_positions_per_second": 28449.053741948843,
+      "required_pages_per_second_upper_estimate": 0.35324203461699394
     }
   ],
   "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": 0.15147967911720173,
-    "producer_p95_seconds": 0.2591939209996781,
-    "gpu_update_seconds": 0.9732003038668886,
-    "producer_batches_per_second": 6.601545539493039,
-    "gpu_batches_per_second": 1.0275376980736917,
-    "producer_headroom": 6.424626125025729,
+    "producer_batch_seconds": 0.18614045605471574,
+    "producer_p95_seconds": 0.23759312900074292,
+    "gpu_update_seconds": 0.9718425171812421,
+    "producer_batches_per_second": 5.372287256597519,
+    "gpu_batches_per_second": 1.0289732979582191,
+    "producer_headroom": 5.221017170472443,
     "sustainable": true,
-    "buffer_seconds": 60.33841883974709,
+    "buffer_seconds": 60.254236065237016,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 443595.31889399345,
-    "upload_bytes_per_second": 4655238.699287984,
+    "generation_bytes_per_second": 452004.44320623693,
+    "upload_bytes_per_second": 4755147.224722894,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.18522906412937,
+    "estimated_full_shards_per_hour": 96.98963138714153,
     "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": 1.090632098000242,
-      "bytes": 1122627,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.176693084999897,
       "bytes": 12183576,
@@ -562,6 +556,12 @@
       "bytes": 1216081,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.5031630230005248,
+      "bytes": 11828419,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 250500, "loss": 0.5950481034815311, "lr": 0.0003, "grad_norm": 0.13582030029360265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36666699586, "tokens_per_second": 148404.6526536854, "data_wait_s": 1.2508999134297483e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250600, "loss": 0.595927158370614, "lr": 0.0003, "grad_norm": 0.15122073928134755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36681307733, "tokens_per_second": 149336.46746914863, "data_wait_s": 1.5302000974770635e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250700, "loss": 0.5758683197200298, "lr": 0.0003, "grad_norm": 0.14111654771792698, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36695727555, "tokens_per_second": 147699.5653436398, "data_wait_s": 1.1117001122329384e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 250800, "loss": 0.6128380689769983, "lr": 0.0003, "grad_norm": 0.1576773410874419, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36710345453, "tokens_per_second": 149630.99916733598, "data_wait_s": 1.034299930324778e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b19515bedf2fd12351a0b99b171a2a7bdb0eea98f3de0b663a422f2f10fe157
-size 290562
+oid sha256:1aa54af3d1e94f67c5243e33cb324a1b49790156db009547b92891d1bdcb029f
+size 292680