Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77327.56690843176,
-      "required_pages_per_second_upper_estimate": 0.005349700503400433
+      "required_token_positions_per_second": 76234.96400137368,
+      "required_pages_per_second_upper_estimate": 0.005274111699102129
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9665.94586355397,
-      "required_pages_per_second_upper_estimate": 0.004046929685167526
+      "required_token_positions_per_second": 9529.37050017171,
+      "required_pages_per_second_upper_estimate": 0.003989748432537275
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19331.89172710794,
-      "required_pages_per_second_upper_estimate": 0.004026323230488081
+      "required_token_positions_per_second": 19058.74100034342,
+      "required_pages_per_second_upper_estimate": 0.003969433137572109
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28997.837590661908,
-      "required_pages_per_second_upper_estimate": 0.0017383760864259096
+      "required_token_positions_per_second": 28588.11150051513,
+      "required_pages_per_second_upper_estimate": 0.0017138136329371257
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28997.837590661908,
-      "required_pages_per_second_upper_estimate": 0.002254454324666091
+      "required_token_positions_per_second": 28588.11150051513,
+      "required_pages_per_second_upper_estimate": 0.002222599923351788
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28997.837590661908,
-      "required_pages_per_second_upper_estimate": 0.37195312516080997
+      "required_token_positions_per_second": 28588.11150051513,
+      "required_pages_per_second_upper_estimate": 0.3666975988060072
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1331795151408528,
-    "producer_p95_seconds": 0.3114297770007397,
-    "gpu_update_seconds": 0.9534504051744674,
-    "producer_batches_per_second": 7.508662266433272,
-    "gpu_batches_per_second": 1.048822250819658,
-    "producer_headroom": 7.159137080249038,
+    "producer_batch_seconds": 0.11150594014043236,
+    "producer_p95_seconds": 0.34420684300130233,
+    "gpu_update_seconds": 0.967115298941723,
+    "producer_batches_per_second": 8.968132090008694,
+    "gpu_batches_per_second": 1.0340028754526596,
+    "producer_headroom": 8.673217747177617,
     "sustainable": true,
-    "buffer_seconds": 60.06737552599145,
+    "buffer_seconds": 60.92826383332855,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450349.58844739257,
-    "upload_bytes_per_second": 5079199.718115679,
+    "generation_bytes_per_second": 392775.76761425514,
+    "upload_bytes_per_second": 5018143.6987694185,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.63453807894071,
+    "estimated_full_shards_per_hour": 84.2805363781046,
     "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": 1.113636714999302,
-      "bytes": 5851090,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.4330363720000605,
       "bytes": 7577948,
@@ -562,6 +556,12 @@
       "bytes": 12216092,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.1287659260051441,
+      "bytes": 715029,
+      "pages": 21,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -349,3 +349,4 @@
 {"step": 639300, "loss": 0.5429550409317017, "lr": 0.0001, "grad_norm": 0.1663941566307602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93577285796, "tokens_per_second": 151397.99076137244, "data_wait_s": 1.624900323804468e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 639400, "loss": 0.5889376401901245, "lr": 0.0001, "grad_norm": 0.18591583123044766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93592017464, "tokens_per_second": 152868.0829272865, "data_wait_s": 1.2397998943924904e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 639500, "loss": 0.5777553915977478, "lr": 0.0001, "grad_norm": 0.17600662552134663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93606628231, "tokens_per_second": 142253.00815013997, "data_wait_s": 1.0637006198521703e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 639600, "loss": 0.5480508208274841, "lr": 0.0001, "grad_norm": 0.17035600415695004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93621211222, "tokens_per_second": 151809.97375823016, "data_wait_s": 1.2152995623182505e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eff42e733c0afc7653ea3f3ae0b9d63cac0efd440d0e2b89d7b04ad71206f28f
-size 749064
+oid sha256:e450bfa0fc91afa36722de1b17315a2dca9219b79493e2897a7f44088feebd68
+size 751182