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": 75927.59890163333,
-      "required_pages_per_second_upper_estimate": 0.006581211836053503
+      "required_token_positions_per_second": 75072.61746135913,
+      "required_pages_per_second_upper_estimate": 0.006507104211741185
     },
     {
       "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": 9490.949862704167,
-      "required_pages_per_second_upper_estimate": 0.0037860818025786526
+      "required_token_positions_per_second": 9384.077182669891,
+      "required_pages_per_second_upper_estimate": 0.0037434486926240193
     },
     {
       "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": 18981.899725408333,
-      "required_pages_per_second_upper_estimate": 0.004391128431447959
+      "required_token_positions_per_second": 18768.154365339782,
+      "required_pages_per_second_upper_estimate": 0.004341682204185956
     },
     {
       "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": 28472.8495881125,
-      "required_pages_per_second_upper_estimate": 0.001733847154419382
+      "required_token_positions_per_second": 28152.231548009677,
+      "required_pages_per_second_upper_estimate": 0.0017143231979299728
     },
     {
       "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": 28472.8495881125,
-      "required_pages_per_second_upper_estimate": 0.0021919187791851298
+      "required_token_positions_per_second": 28152.231548009677,
+      "required_pages_per_second_upper_estimate": 0.002167236714930475
     },
     {
       "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": 28472.8495881125,
-      "required_pages_per_second_upper_estimate": 0.35353749938677254
+      "required_token_positions_per_second": 28152.231548009677,
+      "required_pages_per_second_upper_estimate": 0.3495564963682491
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "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,
+    "producer_batch_seconds": 0.09830264435149161,
+    "producer_p95_seconds": 0.30921685499924934,
+    "gpu_update_seconds": 0.982089109094255,
+    "producer_batches_per_second": 10.172666326496703,
+    "gpu_batches_per_second": 1.0182375415223406,
+    "producer_headroom": 9.990464809702274,
     "sustainable": true,
-    "buffer_seconds": 60.20387930246622,
+    "buffer_seconds": 60.88952476384381,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336206.0774831543,
+    "generation_bytes_per_second": 315997.30765706854,
     "upload_bytes_per_second": 4453940.152774442,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.14199775095913,
+    "estimated_full_shards_per_hour": 67.80566618236581,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -263,3 +263,4 @@
 {"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}
+{"step": 263700, "loss": 0.5276077426970005, "lr": 0.0003, "grad_norm": 0.14215799185759048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38598230894, "tokens_per_second": 147270.508759053, "data_wait_s": 1.5776997315697372e-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 sha256:43d7af9dd43f99a9d79b826c7725e31521db054baf1b7cf8a97e9ceaff8ff81d
-size 565350
+oid sha256:0b8d1dd0f6de79d31c10b6780842539bcf08d40d35c422a2a2274faa11d029aa
+size 567468