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": 74511.30354504212,
-      "required_pages_per_second_upper_estimate": 0.006458450944111941
+      "required_token_positions_per_second": 75927.59890163333,
+      "required_pages_per_second_upper_estimate": 0.006581211836053503
     },
     {
       "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": 9313.912943130264,
-      "required_pages_per_second_upper_estimate": 0.003715459128422796
+      "required_token_positions_per_second": 9490.949862704167,
+      "required_pages_per_second_upper_estimate": 0.0037860818025786526
     },
     {
       "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": 18627.82588626053,
-      "required_pages_per_second_upper_estimate": 0.0043092196802478515
+      "required_token_positions_per_second": 18981.899725408333,
+      "required_pages_per_second_upper_estimate": 0.004391128431447959
     },
     {
       "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": 27941.738829390793,
-      "required_pages_per_second_upper_estimate": 0.001701505295736027
+      "required_token_positions_per_second": 28472.8495881125,
+      "required_pages_per_second_upper_estimate": 0.001733847154419382
     },
     {
       "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": 27941.738829390793,
-      "required_pages_per_second_upper_estimate": 0.002151032402770055
+      "required_token_positions_per_second": 28472.8495881125,
+      "required_pages_per_second_upper_estimate": 0.0021919187791851298
     },
     {
       "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": 27941.738829390793,
-      "required_pages_per_second_upper_estimate": 0.3469428812768143
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28472.8495881125,
+      "required_pages_per_second_upper_estimate": 0.35353749938677254
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2705193206953993,
-    "producer_p95_seconds": 0.31471808400237933,
-    "gpu_update_seconds": 0.9894874534765238,
-    "producer_batches_per_second": 3.69659363859628,
-    "gpu_batches_per_second": 1.0106242342806275,
-    "producer_headroom": 3.65773302599215,
+    "producer_batch_seconds": 0.10336949592186784,
+    "producer_p95_seconds": 0.32851630699951784,
+    "gpu_update_seconds": 0.9710303113301003,
+    "producer_batches_per_second": 9.674033824793469,
+    "gpu_batches_per_second": 1.0298339694774488,
+    "producer_headroom": 9.393780076707124,
     "sustainable": true,
-    "buffer_seconds": 60.35873466206795,
+    "buffer_seconds": 60.20387930246622,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336132.36069043184,
-    "upload_bytes_per_second": 4491860.969745194,
+    "generation_bytes_per_second": 336206.0774831543,
+    "upload_bytes_per_second": 4453940.152774442,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.12617984328,
+    "estimated_full_shards_per_hour": 72.14199775095913,
     "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": 1.1696365809984854,
-      "bytes": 4882259,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.236764369001321,
       "bytes": 13143755,
@@ -562,6 +556,12 @@
       "bytes": 13325077,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.0711962380009936,
+      "bytes": 889212,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 263300, "loss": 0.6413752101361752, "lr": 0.0003, "grad_norm": 0.16248475905404894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38539889874, "tokens_per_second": 149717.19177236749, "data_wait_s": 1.1845997505588457e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 263400, "loss": 0.6026430148631334, "lr": 0.0003, "grad_norm": 0.14385511356056555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38554666149, "tokens_per_second": 151194.80923943364, "data_wait_s": 9.567997039994225e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 263500, "loss": 0.6052782759070396, "lr": 0.0003, "grad_norm": 0.15047077296006206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38569183182, "tokens_per_second": 147396.54202095178, "data_wait_s": 1.3870998373022303e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 263600, "loss": 0.5771237872540951, "lr": 0.0003, "grad_norm": 0.14840465484872778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38583737827, "tokens_per_second": 144652.6355772175, "data_wait_s": 1.4798999472986907e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4196331a588626cf95b226740354122514bf761d013388285c45740fd8ab3839
-size 563232
+oid sha256:43d7af9dd43f99a9d79b826c7725e31521db054baf1b7cf8a97e9ceaff8ff81d
+size 565350