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": 76113.84498009724,
-      "required_pages_per_second_upper_estimate": 0.006597355174098385
+      "required_token_positions_per_second": 76114.12340531524,
+      "required_pages_per_second_upper_estimate": 0.006597379307290619
     },
     {
       "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": 9514.230622512156,
-      "required_pages_per_second_upper_estimate": 0.0037953688457444375
+      "required_token_positions_per_second": 9514.265425664405,
+      "required_pages_per_second_upper_estimate": 0.0037953827292422234
     },
     {
       "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": 19028.46124502431,
-      "required_pages_per_second_upper_estimate": 0.0044018996195563575
+      "required_token_positions_per_second": 19028.53085132881,
+      "required_pages_per_second_upper_estimate": 0.0044019157217498744
     },
     {
       "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": 28542.69186753647,
-      "required_pages_per_second_upper_estimate": 0.001738100182802181
+      "required_token_positions_per_second": 28542.796276993216,
+      "required_pages_per_second_upper_estimate": 0.0017381065407903028
     },
     {
       "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": 28542.69186753647,
-      "required_pages_per_second_upper_estimate": 0.00219729543119099
+      "required_token_positions_per_second": 28542.796276993216,
+      "required_pages_per_second_upper_estimate": 0.0021973034689199947
     },
     {
       "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": 28542.69186753647,
-      "required_pages_per_second_upper_estimate": 0.3544047067501455
+      "required_token_positions_per_second": 28542.796276993216,
+      "required_pages_per_second_upper_estimate": 0.35440600316616233
     }
   ],
   "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": 0.06570757807813266,
-    "producer_p95_seconds": 0.20408313000007183,
-    "gpu_update_seconds": 0.9686542575700766,
-    "producer_batches_per_second": 15.21894474349524,
-    "gpu_batches_per_second": 1.0323600935885586,
-    "producer_headroom": 14.741895621510402,
+    "producer_batch_seconds": 0.10038323982799113,
+    "producer_p95_seconds": 0.2011939269996219,
+    "gpu_update_seconds": 0.9686507142359257,
+    "producer_batches_per_second": 9.961822329240636,
+    "gpu_batches_per_second": 1.0323638699722661,
+    "producer_headroom": 9.649526314310336,
     "sustainable": true,
-    "buffer_seconds": 60.05656396934475,
+    "buffer_seconds": 60.05634428262739,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 519815.6150123726,
-    "upload_bytes_per_second": 4660051.774296162,
+    "generation_bytes_per_second": 519438.2476136463,
+    "upload_bytes_per_second": 4658432.958918611,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.54033029344924,
+    "estimated_full_shards_per_hour": 111.45935603434603,
     "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": 1.1006291130006502,
-      "bytes": 6137509,
-      "pages": 39,
-      "shards": 2
-    },
     {
       "seconds": 1.0797216399996614,
       "bytes": 5987959,
@@ -562,6 +556,12 @@
       "bytes": 5782255,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4370965530015383,
+      "bytes": 7562097,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -138,3 +138,4 @@
 {"step": 250900, "loss": 0.6041381638497114, "lr": 0.0003, "grad_norm": 0.1502225315899803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36725101635, "tokens_per_second": 150323.68269575783, "data_wait_s": 1.3262999345897697e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 251000, "loss": 0.5999299883842468, "lr": 0.0003, "grad_norm": 0.15063485009886873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36739706017, "tokens_per_second": 149506.29546070055, "data_wait_s": 1.5731999155832455e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 251100, "loss": 0.5362557414919138, "lr": 0.0003, "grad_norm": 0.14140154783670644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36754252916, "tokens_per_second": 148980.76062488835, "data_wait_s": 1.633200008654967e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 251200, "loss": 0.5882185809314251, "lr": 0.0003, "grad_norm": 0.17080860579262006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36768800713, "tokens_per_second": 149108.2369015129, "data_wait_s": 1.3812001270707697e-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:f45ba203fee90d99d70859d923687d5b16e97231a1fcf45cc5f9a35791a3d37c
-size 299034
+oid sha256:abb21a6654af9dd8dc5495b0be8f223798ad4267110b5b8974dd793e2f87e76f
+size 301152