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": 76015.50915548955,
-      "required_pages_per_second_upper_estimate": 0.006588831674051258
+      "required_token_positions_per_second": 75554.70956847946,
+      "required_pages_per_second_upper_estimate": 0.0065488907337351
     },
     {
       "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": 9501.938644436193,
-      "required_pages_per_second_upper_estimate": 0.0037904653919084864
+      "required_token_positions_per_second": 9444.338696059933,
+      "required_pages_per_second_upper_estimate": 0.003767487911305223
     },
     {
       "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": 19003.877288872387,
-      "required_pages_per_second_upper_estimate": 0.004396212553963461
+      "required_token_positions_per_second": 18888.677392119866,
+      "required_pages_per_second_upper_estimate": 0.004369563085298702
     },
     {
       "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": 28505.81593330858,
-      "required_pages_per_second_upper_estimate": 0.001735854631880777
+      "required_token_positions_per_second": 28333.016088179796,
+      "required_pages_per_second_upper_estimate": 0.0017253320279231549
     },
     {
       "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": 28505.81593330858,
-      "required_pages_per_second_upper_estimate": 0.002194456619695008
+      "required_token_positions_per_second": 28333.016088179796,
+      "required_pages_per_second_upper_estimate": 0.002181154009276408
     },
     {
       "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": 28505.81593330858,
-      "required_pages_per_second_upper_estimate": 0.35394683106284786
+      "required_token_positions_per_second": 28333.016088179796,
+      "required_pages_per_second_upper_estimate": 0.35180123531022756
     }
   ],
   "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": 0.06198153821866015,
-    "producer_p95_seconds": 0.20213822100049583,
-    "gpu_update_seconds": 0.969907336267255,
-    "producer_batches_per_second": 16.13383644129923,
-    "gpu_batches_per_second": 1.0310263286063577,
-    "producer_headroom": 15.648326326552104,
+    "producer_batch_seconds": 0.11278597261717493,
+    "producer_p95_seconds": 0.19575382999937574,
+    "gpu_update_seconds": 0.9758226908830374,
+    "producer_batches_per_second": 8.866350812917679,
+    "gpu_batches_per_second": 1.024776334207892,
+    "producer_headroom": 8.651986308574335,
     "sustainable": true,
-    "buffer_seconds": 60.13425484856981,
+    "buffer_seconds": 60.50100683474832,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357392.9728841179,
+    "generation_bytes_per_second": 410567.84626226994,
     "upload_bytes_per_second": 5177116.884747958,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.68821229832318,
+    "estimated_full_shards_per_hour": 88.09830227757524,
     "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": [

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -110,3 +110,4 @@
 {"step": 166000, "loss": 0.5583077687770128, "lr": 0.0003, "grad_norm": 0.14372924377867893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24294726970, "tokens_per_second": 148583.72982506774, "data_wait_s": 1.3250000847619958e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 166100, "loss": 0.638989269733429, "lr": 0.0003, "grad_norm": 0.14559203971628143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24309405732, "tokens_per_second": 150312.51889479955, "data_wait_s": 1.1656000424409285e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 166200, "loss": 0.5841880142688751, "lr": 0.0003, "grad_norm": 0.13750225255851747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24324120811, "tokens_per_second": 150169.61155500595, "data_wait_s": 1.296600021305494e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 166300, "loss": 0.5640976019203663, "lr": 0.0003, "grad_norm": 0.1526776613042185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24338797713, "tokens_per_second": 150459.1665480727, "data_wait_s": 1.2185000741737895e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac802c2535f6f3aea4a5dd5b1d73d496b53b306e6d89caebb2da3a6d0fb86074
-size 238947
+oid sha256:a84493c03ffa3d0a4d76bf7c3a9f0abf9356c30faa6be3773ad5f0a976e91414
+size 241065