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": 75313.43075868371,
-      "required_pages_per_second_upper_estimate": 0.006527977298017558
+      "required_token_positions_per_second": 75321.34505531802,
+      "required_pages_per_second_upper_estimate": 0.0065286632891912175
     },
     {
       "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": 9414.178844835464,
-      "required_pages_per_second_upper_estimate": 0.003755456695721822
+      "required_token_positions_per_second": 9415.168131914752,
+      "required_pages_per_second_upper_estimate": 0.00375585133712891
     },
     {
       "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": 18828.357689670927,
-      "required_pages_per_second_upper_estimate": 0.004355609183727646
+      "required_token_positions_per_second": 18830.336263829504,
+      "required_pages_per_second_upper_estimate": 0.004356066892037516
     },
     {
       "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": 28242.53653450639,
-      "required_pages_per_second_upper_estimate": 0.0017198222978139753
+      "required_token_positions_per_second": 28245.504395744254,
+      "required_pages_per_second_upper_estimate": 0.0017200030249921966
     },
     {
       "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": 28242.53653450639,
-      "required_pages_per_second_upper_estimate": 0.0021741886427711944
+      "required_token_positions_per_second": 28245.504395744254,
+      "required_pages_per_second_upper_estimate": 0.0021744171169448506
     },
     {
       "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": 28242.53653450639,
-      "required_pages_per_second_upper_estimate": 0.3506777820691904
+      "required_token_positions_per_second": 28245.504395744254,
+      "required_pages_per_second_upper_estimate": 0.3507146329729721
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.21074995446855382,
-    "producer_p95_seconds": 0.3686122989965952,
-    "gpu_update_seconds": 0.9789488973917058,
-    "producer_batches_per_second": 4.744959506737216,
-    "gpu_batches_per_second": 1.0215037809066259,
-    "producer_headroom": 4.6450728772886904,
+    "producer_batch_seconds": 0.13973798480515143,
+    "producer_p95_seconds": 0.37110723000660073,
+    "gpu_update_seconds": 0.978846035554094,
+    "producer_batches_per_second": 7.15625033089167,
+    "gpu_batches_per_second": 1.0216111254247777,
+    "producer_headroom": 7.004867265825984,
     "sustainable": true,
-    "buffer_seconds": 60.694831638285756,
+    "buffer_seconds": 60.688454204353825,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 446482.51436808065,
-    "upload_bytes_per_second": 5942947.754893314,
+    "generation_bytes_per_second": 444929.1185454948,
+    "upload_bytes_per_second": 6001683.48335616,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.80475400239767,
+    "estimated_full_shards_per_hour": 95.47143142007478,
     "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": 1.0350117630005116,
-      "bytes": 6519406,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.0306341840005189,
       "bytes": 6113692,
@@ -562,6 +556,12 @@
       "bytes": 874907,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.313098856000579,
+      "bytes": 12314988,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -303,3 +303,4 @@
 {"step": 115300, "loss": 0.6172855384647846, "lr": 0.0003, "grad_norm": 0.12460532216136234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16872235692, "tokens_per_second": 148134.06387533378, "data_wait_s": 1.2730000889860094e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 115400, "loss": 0.6104896310716867, "lr": 0.0003, "grad_norm": 0.1571173659230503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16886971989, "tokens_per_second": 149986.69570470662, "data_wait_s": 1.5436999092344195e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 115500, "loss": 0.5834883004426956, "lr": 0.0003, "grad_norm": 0.1572690547827181, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16901693141, "tokens_per_second": 149983.04633133605, "data_wait_s": 1.4511999324895442e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 115600, "loss": 0.7221038471907377, "lr": 0.0003, "grad_norm": 0.17080569046466865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16916387672, "tokens_per_second": 147272.63591024862, "data_wait_s": 7.412003469653428e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7cb5430fb1b28f2752f85c13a81595011f59c30c3c9099a4e1238a17670c78cb
-size 650853
+oid sha256:41fdf92a765c2a5aa1cc0cbb76d3b6fce14f3ea9c5715454bdb8b02c5f74c749
+size 652971