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": 75438.82107633346,
-      "required_pages_per_second_upper_estimate": 0.0065388458129525835
+      "required_token_positions_per_second": 75127.41511690615,
+      "required_pages_per_second_upper_estimate": 0.0065118539336406125
     },
     {
       "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": 9429.852634541683,
-      "required_pages_per_second_upper_estimate": 0.0037617092047796726
+      "required_token_positions_per_second": 9390.926889613269,
+      "required_pages_per_second_upper_estimate": 0.003746181143135978
     },
     {
       "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": 18859.705269083366,
-      "required_pages_per_second_upper_estimate": 0.0043628608945261585
+      "required_token_positions_per_second": 18781.853779226538,
+      "required_pages_per_second_upper_estimate": 0.0043448513224872065
     },
     {
       "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": 28289.55790362505,
-      "required_pages_per_second_upper_estimate": 0.0017226856524912438
+      "required_token_positions_per_second": 28172.78066883981,
+      "required_pages_per_second_upper_estimate": 0.001715574531575623
     },
     {
       "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": 28289.55790362505,
-      "required_pages_per_second_upper_estimate": 0.002177808477929429
+      "required_token_positions_per_second": 28172.78066883981,
+      "required_pages_per_second_upper_estimate": 0.002168818643135586
     },
     {
       "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": 28289.55790362505,
-      "required_pages_per_second_upper_estimate": 0.35126163010324507
+      "required_token_positions_per_second": 28172.78066883981,
+      "required_pages_per_second_upper_estimate": 0.34981164767547596
     }
   ],
   "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": 0.06051090237498613,
-    "producer_p95_seconds": 0.20444957800100383,
-    "gpu_update_seconds": 0.9773217416189159,
-    "producer_batches_per_second": 16.52594756896863,
-    "gpu_batches_per_second": 1.0232044959355124,
-    "producer_headroom": 16.15116786000731,
+    "producer_batch_seconds": 0.08352705639069313,
+    "producer_p95_seconds": 0.26216529000157607,
+    "gpu_update_seconds": 0.9813727769719147,
+    "producer_batches_per_second": 11.972168578796264,
+    "gpu_batches_per_second": 1.0189807822931065,
+    "producer_headroom": 11.749160324549191,
     "sustainable": true,
-    "buffer_seconds": 60.59394798037279,
+    "buffer_seconds": 60.84511217225871,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418637.5580246913,
-    "upload_bytes_per_second": 6351076.292682209,
+    "generation_bytes_per_second": 419675.3027315758,
+    "upload_bytes_per_second": 6287240.16166785,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.82987456851534,
+    "estimated_full_shards_per_hour": 90.05255042515235,
     "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": 1.0477722719988378,
-      "bytes": 4809637,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1933620469990274,
       "bytes": 7601419,
@@ -562,6 +556,12 @@
       "bytes": 7444287,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0936733489998005,
+      "bytes": 876493,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -158,3 +158,4 @@
 {"step": 55800, "loss": 0.6200186684727669, "lr": 0.0003, "grad_norm": 0.13037454080661143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8167305941, "tokens_per_second": 150812.42841196473, "data_wait_s": 1.304299985349644e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 55900, "loss": 0.6394687257707119, "lr": 0.0003, "grad_norm": 0.1317398041031941, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8181978292, "tokens_per_second": 150224.8637059052, "data_wait_s": 1.2283999240025878e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 56000, "loss": 0.6399248819798231, "lr": 0.0003, "grad_norm": 0.13904447677560522, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8196659773, "tokens_per_second": 150288.93443164893, "data_wait_s": 9.495999620412476e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 56100, "loss": 0.6293352022767067, "lr": 0.0003, "grad_norm": 0.12303810071863236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8211177499, "tokens_per_second": 147823.8954392136, "data_wait_s": 1.2841002899222076e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:941b2d2c378b89524a5195f1ffa043d924f4945b9879f23b97711449a551ce85
-size 341394
+oid sha256:7d7ef21df9ec621bf2e0dacee190b6fcc4bbd55043a9c4c0ea4d273f4e96c343
+size 343512