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": 75804.4715616428,
-      "required_pages_per_second_upper_estimate": 0.006570539470286514
+      "required_token_positions_per_second": 76440.67645241093,
+      "required_pages_per_second_upper_estimate": 0.006625684097771742
     },
     {
       "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": 9475.55894520535,
-      "required_pages_per_second_upper_estimate": 0.0037799421354736515
+      "required_token_positions_per_second": 9555.084556551366,
+      "required_pages_per_second_upper_estimate": 0.003811666090853425
     },
     {
       "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": 18951.1178904107,
-      "required_pages_per_second_upper_estimate": 0.00438400759566305
+      "required_token_positions_per_second": 19110.169113102733,
+      "required_pages_per_second_upper_estimate": 0.0044208012968269365
     },
     {
       "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": 28426.67683561605,
-      "required_pages_per_second_upper_estimate": 0.0017310354760420583
+      "required_token_positions_per_second": 28665.253669654096,
+      "required_pages_per_second_upper_estimate": 0.0017455635535190649
     },
     {
       "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": 28426.67683561605,
-      "required_pages_per_second_upper_estimate": 0.0021883642728835937
+      "required_token_positions_per_second": 28665.253669654096,
+      "required_pages_per_second_upper_estimate": 0.002206730577990788
     },
     {
       "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": 28426.67683561605,
-      "required_pages_per_second_upper_estimate": 0.3529641883310286
+      "required_token_positions_per_second": 28665.253669654096,
+      "required_pages_per_second_upper_estimate": 0.35592651414448134
     }
   ],
   "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": 690670,
-    "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,
+    "producer_batch_seconds": 0.06479070506273388,
+    "producer_p95_seconds": 0.20464348400128074,
+    "gpu_update_seconds": 0.9645126576803682,
+    "producer_batches_per_second": 15.43431266925936,
+    "gpu_batches_per_second": 1.036793029139688,
+    "producer_headroom": 14.886589932097122,
     "sustainable": true,
-    "buffer_seconds": 60.30166698389073,
+    "buffer_seconds": 60.7642974338632,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 485921.89544150466,
-    "upload_bytes_per_second": 5357066.625336519,
+    "generation_bytes_per_second": 488330.43009571533,
+    "upload_bytes_per_second": 5301911.8752229875,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.2675270789514,
+    "estimated_full_shards_per_hour": 104.7843425479278,
     "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": 1.2981188600006135,
-      "bytes": 10611352,
-      "pages": 45,
-      "shards": 3
-    },
     {
       "seconds": 1.1970796210007393,
       "bytes": 2312155,
@@ -562,6 +556,12 @@
       "bytes": 2575859,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.209377144001337,
+      "bytes": 5838608,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -105,3 +105,4 @@
 {"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}
+{"step": 291000, "loss": 0.6449759714305401, "lr": 0.0003, "grad_norm": 0.1641612940546156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42594212353, "tokens_per_second": 150595.54317226744, "data_wait_s": 1.2327996955718845e-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:a6f8eaac73f8fe72785e866eea20f2b2278be523b49b5dc9d767b17d4490f48a
-size 228357
+oid sha256:3eae0d7ced578d1670288435f073052c588025ca264de3a902244f59237b9fdf
+size 230475