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": 75971.8592555506,
-      "required_pages_per_second_upper_estimate": 0.006585048211354211
+      "required_token_positions_per_second": 75804.4715616428,
+      "required_pages_per_second_upper_estimate": 0.006570539470286514
     },
     {
       "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": 9496.482406943825,
-      "required_pages_per_second_upper_estimate": 0.0037882888171947605
+      "required_token_positions_per_second": 9475.55894520535,
+      "required_pages_per_second_upper_estimate": 0.0037799421354736515
     },
     {
       "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": 18992.96481388765,
-      "required_pages_per_second_upper_estimate": 0.004393688144928743
+      "required_token_positions_per_second": 18951.1178904107,
+      "required_pages_per_second_upper_estimate": 0.00438400759566305
     },
     {
       "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": 28489.447220831476,
-      "required_pages_per_second_upper_estimate": 0.0017348578631709187
+      "required_token_positions_per_second": 28426.67683561605,
+      "required_pages_per_second_upper_estimate": 0.0017310354760420583
     },
     {
       "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": 28489.447220831476,
-      "required_pages_per_second_upper_estimate": 0.0021931965108970245
+      "required_token_positions_per_second": 28426.67683561605,
+      "required_pages_per_second_upper_estimate": 0.0021883642728835937
     },
     {
       "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": 28489.447220831476,
-      "required_pages_per_second_upper_estimate": 0.3537435864364389
+      "required_token_positions_per_second": 28426.67683561605,
+      "required_pages_per_second_upper_estimate": 0.3529641883310286
     }
   ],
   "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": 0.0968696881327844,
-    "producer_p95_seconds": 0.20495424299951992,
-    "gpu_update_seconds": 0.9704645999513739,
-    "producer_batches_per_second": 10.323146685774885,
-    "gpu_batches_per_second": 1.0304342889478977,
-    "producer_headroom": 10.018248418649875,
+    "producer_batch_seconds": 0.0838775433984722,
+    "producer_p95_seconds": 0.21113853399947402,
+    "gpu_update_seconds": 0.9726075319982377,
+    "producer_batches_per_second": 11.922142202584043,
+    "gpu_batches_per_second": 1.0281639480474556,
+    "producer_headroom": 11.5955653037873,
     "sustainable": true,
-    "buffer_seconds": 60.16880519698518,
+    "buffer_seconds": 60.30166698389073,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 463692.299740859,
+    "generation_bytes_per_second": 485921.89544150466,
     "upload_bytes_per_second": 5357066.625336519,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.49757332009628,
+    "estimated_full_shards_per_hour": 104.2675270789514,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -104,3 +104,4 @@
 {"step": 290600, "loss": 0.5969054363667965, "lr": 0.0003, "grad_norm": 0.17551084322181382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42535497260, "tokens_per_second": 149263.4370258846, "data_wait_s": 1.0021001799032092e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 290700, "loss": 0.623919390141964, "lr": 0.0003, "grad_norm": 0.14654368036302592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42550246399, "tokens_per_second": 151182.37926973036, "data_wait_s": 1.2541000614874065e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 290800, "loss": 0.6278765704482794, "lr": 0.0003, "grad_norm": 0.1561798517446671, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42564830823, "tokens_per_second": 150195.36210640217, "data_wait_s": 2.0540999685181305e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 290900, "loss": 0.5860250983387232, "lr": 0.0003, "grad_norm": 0.13762307720115055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42579513576, "tokens_per_second": 151078.19702101997, "data_wait_s": 1.336499917670153e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:caadcdb381c95136276023cf215f05d8aafca94414879ce2fd6dc225d7460b90
-size 226239
+oid sha256:a6f8eaac73f8fe72785e866eea20f2b2278be523b49b5dc9d767b17d4490f48a
+size 228357