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": 74881.87960242505,
-      "required_pages_per_second_upper_estimate": 0.006490571537549462
+      "required_token_positions_per_second": 75586.73113578392,
+      "required_pages_per_second_upper_estimate": 0.0065516662820311305
     },
     {
       "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": 9360.234950303131,
-      "required_pages_per_second_upper_estimate": 0.003733937669659778
+      "required_token_positions_per_second": 9448.34139197299,
+      "required_pages_per_second_upper_estimate": 0.0037690846465505785
     },
     {
       "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": 18720.469900606262,
-      "required_pages_per_second_upper_estimate": 0.0043306512424877725
+      "required_token_positions_per_second": 18896.68278394598,
+      "required_pages_per_second_upper_estimate": 0.00437141499180759
     },
     {
       "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": 28080.70485090939,
-      "required_pages_per_second_upper_estimate": 0.0017099675973481425
+      "required_token_positions_per_second": 28345.024175918974,
+      "required_pages_per_second_upper_estimate": 0.0017260632574649035
     },
     {
       "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": 28080.70485090939,
-      "required_pages_per_second_upper_estimate": 0.002161730391789125
+      "required_token_positions_per_second": 28345.024175918974,
+      "required_pages_per_second_upper_estimate": 0.002182078425111084
     },
     {
       "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": 28080.70485090939,
-      "required_pages_per_second_upper_estimate": 0.3486683741747196
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28345.024175918974,
+      "required_pages_per_second_upper_estimate": 0.3519503355714637
     }
   ],
   "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": 0.25898029113281495,
-    "producer_p95_seconds": 0.17803285800073354,
-    "gpu_update_seconds": 0.9845906698850054,
-    "producer_batches_per_second": 3.861297690360391,
-    "gpu_batches_per_second": 1.015650493739489,
-    "producer_headroom": 3.8017976795773616,
+    "producer_batch_seconds": 0.07033997385152446,
+    "producer_p95_seconds": 0.17801131200030795,
+    "gpu_update_seconds": 0.9754092932998398,
+    "producer_batches_per_second": 14.216667212740616,
+    "gpu_batches_per_second": 1.0252106545109583,
+    "producer_headroom": 13.867069319058327,
     "sustainable": true,
-    "buffer_seconds": 60.06003086298533,
+    "buffer_seconds": 60.47537618459007,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385926.90721396694,
-    "upload_bytes_per_second": 5235226.743210096,
+    "generation_bytes_per_second": 419371.58761128766,
+    "upload_bytes_per_second": 4058615.1326010227,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.81093036951309,
+    "estimated_full_shards_per_hour": 89.98738023046467,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 7141661,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 3.5350073540002995,
+      "bytes": 5274454,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -37,3 +37,4 @@
 {"step": 158700, "loss": 0.5854216739535332, "lr": 0.0003, "grad_norm": 0.13415142926651943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23226643792, "tokens_per_second": 149593.5826521523, "data_wait_s": 1.518199951533461e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 158800, "loss": 0.6244642548263073, "lr": 0.0003, "grad_norm": 0.14388045756002935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23241394389, "tokens_per_second": 149268.78273229898, "data_wait_s": 1.7186999684781767e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 158900, "loss": 0.5830402486026287, "lr": 0.0003, "grad_norm": 0.1417218542578867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23256080379, "tokens_per_second": 148964.90358281686, "data_wait_s": 1.2634000086109154e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 159000, "loss": 0.5832870006561279, "lr": 0.0003, "grad_norm": 0.14141191893168306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23270696232, "tokens_per_second": 149508.56279893636, "data_wait_s": 1.2864999916928355e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f7449e076051613afa8739bba682178ee8dde8fbfa6c195424f9bf8339403486
-size 82767
+oid sha256:0091fdf0ecdd6a32077efd15af602d7d38425a1c444f9f9bdd653bf59862b550
+size 84885