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": 74158.54575259422,
-      "required_pages_per_second_upper_estimate": 0.006427874792719008
+      "required_token_positions_per_second": 75319.5505841328,
+      "required_pages_per_second_upper_estimate": 0.0065285077489769285
     },
     {
       "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": 9269.818219074277,
-      "required_pages_per_second_upper_estimate": 0.003697869083721987
+      "required_token_positions_per_second": 9414.9438230166,
+      "required_pages_per_second_upper_estimate": 0.00375576185695572
     },
     {
       "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": 18539.636438148555,
-      "required_pages_per_second_upper_estimate": 0.004288818603508423
+      "required_token_positions_per_second": 18829.8876460332,
+      "required_pages_per_second_upper_estimate": 0.004355963112205745
     },
     {
       "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": 27809.454657222832,
-      "required_pages_per_second_upper_estimate": 0.0016934498836924114
+      "required_token_positions_per_second": 28244.8314690498,
+      "required_pages_per_second_upper_estimate": 0.001719962047287078
     },
     {
       "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": 27809.454657222832,
-      "required_pages_per_second_upper_estimate": 0.00214084880101054
+      "required_token_positions_per_second": 28244.8314690498,
+      "required_pages_per_second_upper_estimate": 0.002174365313185131
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27809.454657222832,
-      "required_pages_per_second_upper_estimate": 0.3453003545851327
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28244.8314690498,
+      "required_pages_per_second_upper_estimate": 0.35070627747556776
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2725716539529799,
-    "producer_p95_seconds": 0.4019822469999781,
-    "gpu_update_seconds": 0.9941942530260695,
-    "producer_batches_per_second": 3.668760069132154,
-    "gpu_batches_per_second": 1.0058396505071916,
-    "producer_headroom": 3.6474601764627126,
+    "producer_batch_seconds": 0.14667976332037824,
+    "producer_p95_seconds": 0.411268789001042,
+    "gpu_update_seconds": 0.9788693563385642,
+    "producer_batches_per_second": 6.817573040500467,
+    "gpu_batches_per_second": 1.0215867863516277,
+    "producer_headroom": 6.673513333945841,
     "sustainable": true,
-    "buffer_seconds": 60.64584943459024,
+    "buffer_seconds": 60.68990009299098,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417694.63474757737,
+    "generation_bytes_per_second": 417111.51261794404,
     "upload_bytes_per_second": 3779260.144207867,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.62754518337718,
+    "estimated_full_shards_per_hour": 89.5024207487463,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -375,3 +375,4 @@
 {"step": 352500, "loss": 0.5629769563674927, "lr": 0.0003, "grad_norm": 0.14707417422876704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51596780825, "tokens_per_second": 148714.76398987274, "data_wait_s": 1.1992997315246612e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 352600, "loss": 0.6171035524457693, "lr": 0.0003, "grad_norm": 0.1642221486516797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51611234360, "tokens_per_second": 146443.17700148126, "data_wait_s": 1.555299968458712e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 352700, "loss": 0.5836805626749992, "lr": 0.0003, "grad_norm": 0.14487112585091855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51625820172, "tokens_per_second": 148252.52253081976, "data_wait_s": 1.5049998182803392e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 352800, "loss": 0.5335069727152586, "lr": 0.0003, "grad_norm": 0.13403092636276778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51640409496, "tokens_per_second": 139846.56906891658, "data_wait_s": 1.3468001270666718e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b4fa4851611a70e357aab277656cebb19a089331a2323643b2232a50280ae76e
-size 804132
+oid sha256:b4ea6636653995243f3135e9ce1bf7f694e59888d4a162245bcbf3afcc931528
+size 806250