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": 75516.06790801907,
-      "required_pages_per_second_upper_estimate": 0.006545541372542787
+      "required_token_positions_per_second": 75567.44603663964,
+      "required_pages_per_second_upper_estimate": 0.006549994698514941
     },
     {
       "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": 9439.508488502384,
-      "required_pages_per_second_upper_estimate": 0.003765561069292478
+      "required_token_positions_per_second": 9445.930754579955,
+      "required_pages_per_second_upper_estimate": 0.0037681230072522556
     },
     {
       "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": 18879.01697700477,
-      "required_pages_per_second_upper_estimate": 0.004367328318279324
+      "required_token_positions_per_second": 18891.86150915991,
+      "required_pages_per_second_upper_estimate": 0.00437029967473737
     },
     {
       "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": 28318.525465507155,
-      "required_pages_per_second_upper_estimate": 0.0017244496250288126
+      "required_token_positions_per_second": 28337.792263739862,
+      "required_pages_per_second_upper_estimate": 0.0017256228719561065
     },
     {
       "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": 28318.525465507155,
-      "required_pages_per_second_upper_estimate": 0.0021800384810304572
+      "required_token_positions_per_second": 28337.792263739862,
+      "required_pages_per_second_upper_estimate": 0.002181521692492322
     },
     {
       "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": 28318.525465507155,
-      "required_pages_per_second_upper_estimate": 0.3516213102736277
+      "required_token_positions_per_second": 28337.792263739862,
+      "required_pages_per_second_upper_estimate": 0.35186053942585227
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3679920735389999,
-    "producer_p95_seconds": 0.21067002699965087,
-    "gpu_update_seconds": 0.9763220204977171,
-    "producer_batches_per_second": 2.717449836304748,
-    "gpu_batches_per_second": 1.02425222314479,
-    "producer_headroom": 2.653106114782242,
+    "producer_batch_seconds": 0.05354392135159003,
+    "producer_p95_seconds": 0.18385103800028446,
+    "gpu_update_seconds": 0.9756582214549403,
+    "producer_batches_per_second": 18.67625632858704,
+    "gpu_batches_per_second": 1.0249490836132762,
+    "producer_headroom": 18.221643032985803,
     "sustainable": true,
-    "buffer_seconds": 60.53196527085846,
+    "buffer_seconds": 60.4908097302063,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304877.81003685505,
-    "upload_bytes_per_second": 4906323.104060274,
+    "generation_bytes_per_second": 305477.72628137865,
+    "upload_bytes_per_second": 5353574.930468796,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.41968084172477,
+    "estimated_full_shards_per_hour": 65.5484089024641,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 2613270,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1979163849991892,
+      "bytes": 9677162,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -55,3 +55,4 @@
 {"step": 90500, "loss": 0.6523060668259859, "lr": 0.0003, "grad_norm": 0.1484641486469891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13244100807, "tokens_per_second": 150066.975510159, "data_wait_s": 1.3369999578571878e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 90600, "loss": 0.5686690304428339, "lr": 0.0003, "grad_norm": 0.13435895790141134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13258850239, "tokens_per_second": 150498.51317028955, "data_wait_s": 9.449000572203659e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 90700, "loss": 0.5889896973967552, "lr": 0.0003, "grad_norm": 0.13338667831945086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13273557530, "tokens_per_second": 149663.66695066163, "data_wait_s": 1.3901000784244388e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 90800, "loss": 0.6065632402896881, "lr": 0.0003, "grad_norm": 0.1261719028856433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13288072735, "tokens_per_second": 147650.11146710603, "data_wait_s": 9.131999831879511e-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:c5d45323be304adc4b8bb608511a3ed14c313a5f89122e14dfb60b98b4c09bca
-size 121674
+oid sha256:8d1495a1ae636feadf3915037ff88354601e2c6bff4f20822fd2aa57d9293a38
+size 123792