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": 75295.99649964369,
-      "required_pages_per_second_upper_estimate": 0.006526466140630694
+      "required_token_positions_per_second": 75550.19877153687,
+      "required_pages_per_second_upper_estimate": 0.006548499749288627
     },
     {
       "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": 9411.999562455461,
-      "required_pages_per_second_upper_estimate": 0.003754587347397264
+      "required_token_positions_per_second": 9443.77484644211,
+      "required_pages_per_second_upper_estimate": 0.00376726298326237
     },
     {
       "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": 18823.999124910923,
-      "required_pages_per_second_upper_estimate": 0.004354600906478538
+      "required_token_positions_per_second": 18887.54969288422,
+      "required_pages_per_second_upper_estimate": 0.004369302211927364
     },
     {
       "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": 28235.998687366384,
-      "required_pages_per_second_upper_estimate": 0.0017194241772245818
+      "required_token_positions_per_second": 28331.324539326328,
+      "required_pages_per_second_upper_estimate": 0.0017252290214728498
     },
     {
       "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": 28235.998687366384,
-      "required_pages_per_second_upper_estimate": 0.002173685341200437
+      "required_token_positions_per_second": 28331.324539326328,
+      "required_pages_per_second_upper_estimate": 0.002181023789163161
     },
     {
       "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": 28235.998687366384,
-      "required_pages_per_second_upper_estimate": 0.3505966038884784
+      "required_token_positions_per_second": 28331.324539326328,
+      "required_pages_per_second_upper_estimate": 0.3517802319347173
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.08905426654700932,
-    "producer_p95_seconds": 0.32267627499822993,
-    "gpu_update_seconds": 0.9791755661318446,
-    "producer_batches_per_second": 11.22910825919977,
-    "gpu_batches_per_second": 1.02126731363449,
-    "producer_headroom": 10.995268436857705,
+    "producer_batch_seconds": 0.08801387023427765,
+    "producer_p95_seconds": 0.3242520589992637,
+    "gpu_update_seconds": 0.975880953310961,
+    "producer_batches_per_second": 11.361845551595147,
+    "gpu_batches_per_second": 1.0247151526087357,
+    "producer_headroom": 11.087808668262573,
     "sustainable": true,
-    "buffer_seconds": 60.70888510017437,
+    "buffer_seconds": 60.50461910527958,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321234.6627344205,
+    "generation_bytes_per_second": 335339.4595371224,
     "upload_bytes_per_second": 3979676.12432033,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.92948066257917,
+    "estimated_full_shards_per_hour": 71.95604171357397,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -144,3 +144,4 @@
 {"step": 294600, "loss": 0.6203990988433361, "lr": 0.0003, "grad_norm": 0.19392050858719598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43121617038, "tokens_per_second": 150742.44589553765, "data_wait_s": 1.0978001228068024e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 294700, "loss": 0.6058078296482563, "lr": 0.0003, "grad_norm": 0.16915365988143374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43136199010, "tokens_per_second": 148780.71310533409, "data_wait_s": 1.2674001482082531e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 294800, "loss": 0.6485474109649658, "lr": 0.0003, "grad_norm": 0.17541547180358677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43150702227, "tokens_per_second": 147358.1234483947, "data_wait_s": 9.000999853014946e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 294900, "loss": 0.6201039385050535, "lr": 0.0003, "grad_norm": 0.1613886835568409, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43165218484, "tokens_per_second": 148639.20341720953, "data_wait_s": 1.2045999028487131e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0ded3199c8d856e202f86c29bf182073d8713633c46a59942786416c02c1b86
-size 310959
+oid sha256:8e8c784a5785f2dde788b4ff0c5442a1cb7d6288c7ceadf7990610ff50f38f8d
+size 313077