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": 75702.22144260752,
-      "required_pages_per_second_upper_estimate": 0.006561676689119094
+      "required_token_positions_per_second": 76518.91519536075,
+      "required_pages_per_second_upper_estimate": 0.006632465633716351
     },
     {
       "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": 9462.77768032594,
-      "required_pages_per_second_upper_estimate": 0.003774843497816316
+      "required_token_positions_per_second": 9564.864399420094,
+      "required_pages_per_second_upper_estimate": 0.0038155674163954417
     },
     {
       "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": 18925.55536065188,
-      "required_pages_per_second_upper_estimate": 0.004378094154288505
+      "required_token_positions_per_second": 19129.728798840188,
+      "required_pages_per_second_upper_estimate": 0.004425326085883585
     },
     {
       "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": 28388.33304097782,
-      "required_pages_per_second_upper_estimate": 0.0017287005401229333
+      "required_token_positions_per_second": 28694.59319826028,
+      "required_pages_per_second_upper_estimate": 0.0017473501768785716
     },
     {
       "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": 28388.33304097782,
-      "required_pages_per_second_upper_estimate": 0.002185412461429926
+      "required_token_positions_per_second": 28694.59319826028,
+      "required_pages_per_second_upper_estimate": 0.0022089892161199056
     },
     {
       "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": 28388.33304097782,
-      "required_pages_per_second_upper_estimate": 0.35248808673004733
+      "required_token_positions_per_second": 28694.59319826028,
+      "required_pages_per_second_upper_estimate": 0.3562908128966845
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10236522223425482,
-    "producer_p95_seconds": 0.3400788390026719,
-    "gpu_update_seconds": 0.9739212217952646,
-    "producer_batches_per_second": 9.768942793008138,
-    "gpu_batches_per_second": 1.0267770920492556,
-    "producer_headroom": 9.514180700614531,
+    "producer_batch_seconds": 0.11378976910157235,
+    "producer_p95_seconds": 0.3222329300006095,
+    "gpu_update_seconds": 0.9635264667796811,
+    "producer_batches_per_second": 8.788136296395578,
+    "gpu_batches_per_second": 1.0378542100065207,
+    "producer_headroom": 8.467601915244304,
     "sustainable": true,
-    "buffer_seconds": 60.383115751306406,
+    "buffer_seconds": 60.70216740711991,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315063.1909580651,
-    "upload_bytes_per_second": 4974779.8212099625,
+    "generation_bytes_per_second": 315050.71230370016,
+    "upload_bytes_per_second": 5375755.666983535,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.60522648388353,
+    "estimated_full_shards_per_hour": 67.60254885514502,
     "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": [
-    {
-      "seconds": 7.225236703998235,
-      "bytes": 1179968,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 4.945576189998974,
       "bytes": 7777829,
@@ -562,6 +556,12 @@
       "bytes": 6178782,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1377685129991733,
+      "bytes": 2613014,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -243,3 +243,4 @@
 {"step": 304500, "loss": 0.6310973726212978, "lr": 0.0003, "grad_norm": 0.16308966050276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44570812241, "tokens_per_second": 149366.05639460578, "data_wait_s": 1.0029001714428887e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 304600, "loss": 0.5887370780110359, "lr": 0.0003, "grad_norm": 0.1347984702876702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44585499420, "tokens_per_second": 149879.98312117712, "data_wait_s": 1.2803000572603196e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 304700, "loss": 0.5947480089962482, "lr": 0.0003, "grad_norm": 0.14075108484866727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44600248529, "tokens_per_second": 150838.17199795527, "data_wait_s": 1.058699854183942e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 304800, "loss": 0.5849448200315237, "lr": 0.0003, "grad_norm": 0.16570698714419854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44614851586, "tokens_per_second": 143666.1496722523, "data_wait_s": 1.3000997569179162e-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:f90cd0eb3b32168702394a88515a5a836f993852ee01a6959e0bb16387b92e69
-size 522207
+oid sha256:6b3f745593a8051657bfeaad1751ada89777e75d6d37b127d1dcb7b703281dc5
+size 524325