Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76360.43811392518,
-      "required_pages_per_second_upper_estimate": 0.0052827922893484945
+      "required_token_positions_per_second": 75566.73858739756,
+      "required_pages_per_second_upper_estimate": 0.005227882314466684
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9545.054764240647,
-      "required_pages_per_second_upper_estimate": 0.0039963151063782615
+      "required_token_positions_per_second": 9445.842323424695,
+      "required_pages_per_second_upper_estimate": 0.0039547769291999775
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19090.109528481295,
-      "required_pages_per_second_upper_estimate": 0.003975966374739511
+      "required_token_positions_per_second": 18891.68464684939,
+      "required_pages_per_second_upper_estimate": 0.003934639704711605
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28635.16429272194,
-      "required_pages_per_second_upper_estimate": 0.0017166343759844702
+      "required_token_positions_per_second": 28337.526970274084,
+      "required_pages_per_second_upper_estimate": 0.0016987914729695982
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28635.16429272194,
-      "required_pages_per_second_upper_estimate": 0.002226258070982507
+      "required_token_positions_per_second": 28337.526970274084,
+      "required_pages_per_second_upper_estimate": 0.0022031180783304103
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28635.16429272194,
-      "required_pages_per_second_upper_estimate": 0.3673011415030841
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28337.526970274084,
+      "required_pages_per_second_upper_estimate": 0.36348336950878113
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07445362869532346,
-    "producer_p95_seconds": 0.262159181998868,
-    "gpu_update_seconds": 0.9655261522989466,
-    "producer_batches_per_second": 13.431178809190953,
-    "gpu_batches_per_second": 1.0357047270226396,
-    "producer_headroom": 12.968154396477289,
+    "producer_batch_seconds": 0.08918417104672471,
+    "producer_p95_seconds": 0.2558650510000007,
+    "gpu_update_seconds": 0.9756673554824529,
+    "producer_batches_per_second": 11.21275208664649,
+    "gpu_batches_per_second": 1.0249394882188254,
+    "producer_headroom": 10.939916176058736,
     "sustainable": true,
-    "buffer_seconds": 60.82814759483363,
+    "buffer_seconds": 60.491376039912076,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 533912.9312389292,
+    "generation_bytes_per_second": 534651.9461649449,
     "upload_bytes_per_second": 3298392.6802292876,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.56528618694216,
+    "estimated_full_shards_per_hour": 114.72386158667813,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -249,3 +249,4 @@
 {"step": 824900, "loss": 0.5088322162628174, "lr": 3e-05, "grad_norm": 0.29723642837319825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120742427267, "tokens_per_second": 150689.67501326322, "data_wait_s": 1.1973999789915979e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 825000, "loss": 0.5475109219551086, "lr": 3e-05, "grad_norm": 0.29112188574392844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120756968551, "tokens_per_second": 150644.89778821776, "data_wait_s": 9.749997843755409e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 825100, "loss": 0.5449039340019226, "lr": 3e-05, "grad_norm": 0.30113785053369213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120771493381, "tokens_per_second": 140349.16175821345, "data_wait_s": 1.1435997294029221e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 825200, "loss": 0.5410646200180054, "lr": 3e-05, "grad_norm": 0.298901420001251, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120786055572, "tokens_per_second": 150135.7085567645, "data_wait_s": 9.097999281948432e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b0bffcbddfa92b8b86d36fa076c5115c831c7a130edcb8baeb8965ff6c87566a
-size 535698
+oid sha256:91c29fc8b6da9815fa81bb577fbf4bb05a92ab8e4b6873dcdf094700ebd1f9d4
+size 537816