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": 74279.87317241088,
-      "required_pages_per_second_upper_estimate": 0.006438391145967182
+      "required_token_positions_per_second": 74922.61591509571,
+      "required_pages_per_second_upper_estimate": 0.006494102457886515
     },
     {
       "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": 9284.98414655136,
-      "required_pages_per_second_upper_estimate": 0.003703918998943418
+      "required_token_positions_per_second": 9365.326989386964,
+      "required_pages_per_second_upper_estimate": 0.0037359689601830877
     },
     {
       "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": 18569.96829310272,
-      "required_pages_per_second_upper_estimate": 0.004295835344329655
+      "required_token_positions_per_second": 18730.653978773928,
+      "required_pages_per_second_upper_estimate": 0.004333007149738208
     },
     {
       "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": 27854.95243965408,
-      "required_pages_per_second_upper_estimate": 0.0016962204599340583
+      "required_token_positions_per_second": 28095.980968160893,
+      "required_pages_per_second_upper_estimate": 0.001710897832741166
     },
     {
       "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": 27854.95243965408,
-      "required_pages_per_second_upper_estimate": 0.00214435134624802
+      "required_token_positions_per_second": 28095.980968160893,
+      "required_pages_per_second_upper_estimate": 0.002162906389582145
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27854.95243965408,
-      "required_pages_per_second_upper_estimate": 0.3458652847716463
+      "required_token_positions_per_second": 28095.980968160893,
+      "required_pages_per_second_upper_estimate": 0.34885805242510765
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2317790046171737,
-    "producer_p95_seconds": 0.17695051500049885,
-    "gpu_update_seconds": 0.992570353867865,
-    "producer_batches_per_second": 4.314454631694042,
-    "gpu_batches_per_second": 1.007485258957396,
-    "producer_headroom": 4.2823997605274045,
+    "producer_batch_seconds": 0.05082951893749055,
+    "producer_p95_seconds": 0.17748967499937862,
+    "gpu_update_seconds": 0.9840553362892523,
+    "producer_batches_per_second": 19.67360740182858,
+    "gpu_batches_per_second": 1.0162030153414674,
+    "producer_headroom": 19.359918347829147,
     "sustainable": true,
-    "buffer_seconds": 60.546791585939765,
+    "buffer_seconds": 60.027375513644394,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328113.0883383849,
-    "upload_bytes_per_second": 4268388.328580749,
+    "generation_bytes_per_second": 314335.48057731526,
+    "upload_bytes_per_second": 4315165.3297424745,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.40543067563688,
+    "estimated_full_shards_per_hour": 67.44907677640528,
     "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": [
@@ -442,6 +442,12 @@
       "bytes": 11563576,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 0.9312379190005231,
+      "bytes": 7230860,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -48,3 +48,4 @@
 {"step": 4800, "loss": 0.8045641928911209, "lr": 0.0003, "grad_norm": 0.18913845286306885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 702310909, "tokens_per_second": 150303.99248884092, "data_wait_s": 1.3591999959317036e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 4900, "loss": 0.7926367856562138, "lr": 0.0003, "grad_norm": 0.18977786811464437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 716901524, "tokens_per_second": 148439.51503401782, "data_wait_s": 1.1121999705210328e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 5000, "loss": 0.7549786865711212, "lr": 0.0003, "grad_norm": 0.16393758598977393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 731411255, "tokens_per_second": 147063.72191024804, "data_wait_s": 1.298300048802048e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 5100, "loss": 0.7820186764001846, "lr": 0.0003, "grad_norm": 0.18854054384762203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 745997968, "tokens_per_second": 136184.03242160723, "data_wait_s": 1.439300012862077e-05, "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 sha256:104a2844cc05443a22269f6bb141741821ff7ef17ee2d394d153057f3da5685a
-size 105049
+oid sha256:ada4f1910f17f27e56430a232956aaede24a21ccd5b0fca9ecefde1109e70a22
+size 107132