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": 75413.41577504123,
-      "required_pages_per_second_upper_estimate": 0.00653664374582572
+      "required_token_positions_per_second": 75361.11575349608,
+      "required_pages_per_second_upper_estimate": 0.006532110512511379
     },
     {
       "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": 9426.676971880153,
-      "required_pages_per_second_upper_estimate": 0.0037604423854636004
+      "required_token_positions_per_second": 9420.13946918701,
+      "required_pages_per_second_upper_estimate": 0.0037578344778949296
     },
     {
       "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": 18853.353943760307,
-      "required_pages_per_second_upper_estimate": 0.004361391627192177
+      "required_token_positions_per_second": 18840.27893837402,
+      "required_pages_per_second_upper_estimate": 0.004358366954808295
     },
     {
       "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": 28280.030915640462,
-      "required_pages_per_second_upper_estimate": 0.0017221055089072254
+      "required_token_positions_per_second": 28260.41840756103,
+      "required_pages_per_second_upper_estimate": 0.0017209112100640654
     },
     {
       "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": 28280.030915640462,
-      "required_pages_per_second_upper_estimate": 0.002177075064021985
+      "required_token_positions_per_second": 28260.41840756103,
+      "required_pages_per_second_upper_estimate": 0.002175565238859133
     },
     {
       "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": 28280.030915640462,
-      "required_pages_per_second_upper_estimate": 0.3511433367972542
+      "required_token_positions_per_second": 28260.41840756103,
+      "required_pages_per_second_upper_estimate": 0.35089981508574974
     }
   ],
   "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": 0.07465564274227177,
-    "producer_p95_seconds": 0.275424565999856,
-    "gpu_update_seconds": 0.977650982153244,
-    "producer_batches_per_second": 13.394834780971976,
-    "gpu_batches_per_second": 1.0228599144835235,
-    "producer_headroom": 13.095473379397687,
+    "producer_batch_seconds": 0.09893946624217165,
+    "producer_p95_seconds": 0.3258566270014853,
+    "gpu_update_seconds": 0.9783294642446916,
+    "producer_batches_per_second": 10.107190163653451,
+    "gpu_batches_per_second": 1.0221505500419932,
+    "producer_headroom": 9.888161937826299,
     "sustainable": true,
-    "buffer_seconds": 60.61436089350113,
+    "buffer_seconds": 60.65642678317088,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390541.3201552291,
+    "generation_bytes_per_second": 432833.62458750274,
     "upload_bytes_per_second": 5688061.204038365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.80107596867232,
+    "estimated_full_shards_per_hour": 92.87601998537838,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -226,3 +226,4 @@
 {"step": 22600, "loss": 0.6810497716069221, "lr": 0.0003, "grad_norm": 0.13040007951598498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3306563191, "tokens_per_second": 146442.26679563854, "data_wait_s": 1.208100002259016e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 22700, "loss": 0.6697792634367943, "lr": 0.0003, "grad_norm": 0.13214199019849326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3321172444, "tokens_per_second": 148412.0297910993, "data_wait_s": 1.2770000466844067e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 22800, "loss": 0.6651651300489902, "lr": 0.0003, "grad_norm": 0.13975925019841318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3335845366, "tokens_per_second": 149251.86210631623, "data_wait_s": 1.0806001228047535e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 22900, "loss": 0.6729341298341751, "lr": 0.0003, "grad_norm": 0.13222387328376883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3350480304, "tokens_per_second": 148574.3674867066, "data_wait_s": 9.93300272966735e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:24ee2f4219afb07bf44e84ad03cb92971d3cc91ac528313583b514963d445a49
-size 480427
+oid sha256:41d00ad282ff0da0061f00031a309ca41a0bdc9c22e8bfc37d59899d2ca53d1c
+size 482545