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": 74788.76666548502,
-      "required_pages_per_second_upper_estimate": 0.006482500744168082
+      "required_token_positions_per_second": 75102.22448560117,
+      "required_pages_per_second_upper_estimate": 0.0065096704735641674
     },
     {
       "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": 9348.595833185627,
-      "required_pages_per_second_upper_estimate": 0.0037292946518212968
+      "required_token_positions_per_second": 9387.778060700146,
+      "required_pages_per_second_upper_estimate": 0.003744925028203345
     },
     {
       "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": 18697.191666371255,
-      "required_pages_per_second_upper_estimate": 0.0043252662326804334
+      "required_token_positions_per_second": 18775.556121400292,
+      "required_pages_per_second_upper_estimate": 0.004343394470184097
     },
     {
       "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": 28045.787499556878,
-      "required_pages_per_second_upper_estimate": 0.0017078413138479588
+      "required_token_positions_per_second": 28163.334182100436,
+      "required_pages_per_second_upper_estimate": 0.0017149992900950808
     },
     {
       "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": 28045.787499556878,
-      "required_pages_per_second_upper_estimate": 0.0021590423574245937
+      "required_token_positions_per_second": 28163.334182100436,
+      "required_pages_per_second_upper_estimate": 0.00216809142643684
     },
     {
       "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": 28045.787499556878,
-      "required_pages_per_second_upper_estimate": 0.34823481753177893
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28163.334182100436,
+      "required_pages_per_second_upper_estimate": 0.34969435392553033
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.18816752620304555,
-    "producer_p95_seconds": 0.35766000600051484,
-    "gpu_update_seconds": 0.9858164974129122,
-    "producer_batches_per_second": 5.314413279371766,
-    "gpu_batches_per_second": 1.0143875687050377,
-    "producer_headroom": 5.239036284874943,
+    "producer_batch_seconds": 0.14321208446861533,
+    "producer_p95_seconds": 0.37185477900129627,
+    "gpu_update_seconds": 0.9817019469794183,
+    "producer_batches_per_second": 6.982650966295712,
+    "gpu_batches_per_second": 1.0186391124891652,
+    "producer_headroom": 6.854882048690217,
     "sustainable": true,
-    "buffer_seconds": 60.134806342187645,
+    "buffer_seconds": 60.865520712723935,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 249015.1103870371,
+    "generation_bytes_per_second": 269297.7603940833,
     "upload_bytes_per_second": 5295981.193420068,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.43284591396651,
+    "estimated_full_shards_per_hour": 57.78503044955133,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -311,3 +311,4 @@
 {"step": 226100, "loss": 0.6101761758327484, "lr": 0.0003, "grad_norm": 0.14038425550628747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33094931729, "tokens_per_second": 150510.12659799188, "data_wait_s": 1.3609998859465122e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 226200, "loss": 0.5835002064704895, "lr": 0.0003, "grad_norm": 0.13729538664677776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33109675151, "tokens_per_second": 150500.3309403648, "data_wait_s": 1.211600101669319e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 226300, "loss": 0.5875837076455355, "lr": 0.0003, "grad_norm": 0.1494478455169779, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33124224171, "tokens_per_second": 147774.17393377807, "data_wait_s": 1.5002999134594575e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 226400, "loss": 0.5998847484588623, "lr": 0.0003, "grad_norm": 0.14292261352133273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33138742060, "tokens_per_second": 147449.55271050605, "data_wait_s": 2.0502997358562425e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a902ad609c7db8e4db01938d3a143c13c6d06c9ee68c59288bba220937409c44
-size 667797
+oid sha256:4a724808b4e2bbfb38588e89d545962c2029fa4295a7b861f4d12824123a19bd
+size 669915