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": 75733.22023183168,
-      "required_pages_per_second_upper_estimate": 0.006564363585603331
+      "required_token_positions_per_second": 76255.53528100096,
+      "required_pages_per_second_upper_estimate": 0.006609636530269935
     },
     {
       "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": 9466.65252897896,
-      "required_pages_per_second_upper_estimate": 0.0037763892328781558
+      "required_token_positions_per_second": 9531.94191012512,
+      "required_pages_per_second_upper_estimate": 0.003802434143180597
     },
     {
       "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": 18933.30505795792,
-      "required_pages_per_second_upper_estimate": 0.004379886910370243
+      "required_token_positions_per_second": 19063.88382025024,
+      "required_pages_per_second_upper_estimate": 0.0044100940089188455
     },
     {
       "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": 28399.957586936885,
-      "required_pages_per_second_upper_estimate": 0.0017294084139825611
+      "required_token_positions_per_second": 28595.82573037536,
+      "required_pages_per_second_upper_estimate": 0.0017413357562772358
     },
     {
       "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": 28399.957586936885,
-      "required_pages_per_second_upper_estimate": 0.0021863073511566573
+      "required_token_positions_per_second": 28595.82573037536,
+      "required_pages_per_second_upper_estimate": 0.0022013858230362743
     },
     {
       "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": 28399.957586936885,
-      "required_pages_per_second_upper_estimate": 0.35263242468600625
+      "required_token_positions_per_second": 28595.82573037536,
+      "required_pages_per_second_upper_estimate": 0.35506445149900495
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11673562643738933,
-    "producer_p95_seconds": 0.3467992559999402,
-    "gpu_update_seconds": 0.9735225806364316,
-    "producer_batches_per_second": 8.566365132210482,
-    "gpu_batches_per_second": 1.0271975400367797,
-    "producer_headroom": 8.339549890183495,
+    "producer_batch_seconds": 0.10725184135162635,
+    "producer_p95_seconds": 0.322419145002641,
+    "gpu_update_seconds": 0.9668544024812492,
+    "producer_batches_per_second": 9.323849244895376,
+    "gpu_batches_per_second": 1.0342818912896181,
+    "producer_headroom": 9.014804690498565,
     "sustainable": true,
-    "buffer_seconds": 60.35839999945876,
+    "buffer_seconds": 60.9118273563187,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303404.3632787559,
-    "upload_bytes_per_second": 4565500.588017751,
+    "generation_bytes_per_second": 352663.5103043635,
+    "upload_bytes_per_second": 4570978.185577136,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.10351346752175,
+    "estimated_full_shards_per_hour": 75.67337972496203,
     "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": 1.3760789849984576,
-      "bytes": 7774763,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.7174801740002295,
       "bytes": 11887929,
@@ -562,6 +556,12 @@
       "bytes": 5772648,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2891920199981541,
+      "bytes": 7891753,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -305,3 +305,4 @@
 {"step": 185500, "loss": 0.6365750879049301, "lr": 0.0003, "grad_norm": 0.13586671619894747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27149107625, "tokens_per_second": 149273.30140484785, "data_wait_s": 8.655999408802018e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 185600, "loss": 0.6221201792359352, "lr": 0.0003, "grad_norm": 0.13968111211190845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27163702481, "tokens_per_second": 149037.87691108143, "data_wait_s": 1.4326000382425264e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 185700, "loss": 0.6390447355806828, "lr": 0.0003, "grad_norm": 0.17158830601594321, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27178325663, "tokens_per_second": 150389.95777772815, "data_wait_s": 8.137001714203507e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 185800, "loss": 0.5502657704055309, "lr": 0.0003, "grad_norm": 0.14309495935863623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27192935771, "tokens_per_second": 149516.55377683364, "data_wait_s": 1.3199998647905886e-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:b76d0b62039d21133b77c3979960b1ce6c9e19a28b00314a758f9092aec6774b
-size 655089
+oid sha256:9623a03d661e4ab632570285ce53f8d1fbac1413d0493e1b71af1c598229c17b
+size 657207