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": 77192.97813029129,
-      "required_pages_per_second_upper_estimate": 0.005340389339439685
+      "required_token_positions_per_second": 76603.37079774599,
+      "required_pages_per_second_upper_estimate": 0.0052995989360967045
     },
     {
       "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": 9649.122266286411,
-      "required_pages_per_second_upper_estimate": 0.00403988599630826
+      "required_token_positions_per_second": 9575.421349718248,
+      "required_pages_per_second_upper_estimate": 0.004009028961591319
     },
     {
       "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": 19298.244532572822,
-      "required_pages_per_second_upper_estimate": 0.004019315407202607
+      "required_token_positions_per_second": 19150.842699436496,
+      "required_pages_per_second_upper_estimate": 0.003988615492608056
     },
     {
       "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": 28947.36679885923,
-      "required_pages_per_second_upper_estimate": 0.001735350439521776
+      "required_token_positions_per_second": 28726.264049154743,
+      "required_pages_per_second_upper_estimate": 0.001722095667281343
     },
     {
       "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": 28947.36679885923,
-      "required_pages_per_second_upper_estimate": 0.0022505304426009847
+      "required_token_positions_per_second": 28726.264049154743,
+      "required_pages_per_second_upper_estimate": 0.002233340676339677
     },
     {
       "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": 28947.36679885923,
-      "required_pages_per_second_upper_estimate": 0.3713057400348794
+      "required_token_positions_per_second": 28726.264049154743,
+      "required_pages_per_second_upper_estimate": 0.36846967136330655
     }
   ],
   "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": 0.2541575213281817,
-    "producer_p95_seconds": 0.34577259900106583,
-    "gpu_update_seconds": 0.9551127807966824,
-    "producer_batches_per_second": 3.9345678017875647,
-    "gpu_batches_per_second": 1.0469967736855914,
-    "producer_headroom": 3.7579559943984107,
+    "producer_batch_seconds": 0.13780444442977569,
+    "producer_p95_seconds": 0.3173430500028189,
+    "gpu_update_seconds": 0.9624641739938866,
+    "producer_batches_per_second": 7.2566600020625165,
+    "gpu_batches_per_second": 1.038999712426025,
+    "producer_headroom": 6.984275274839575,
     "sustainable": true,
-    "buffer_seconds": 60.17210519019099,
+    "buffer_seconds": 60.63524296161486,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 452541.4094379289,
-    "upload_bytes_per_second": 5013109.881515175,
+    "generation_bytes_per_second": 450225.8879258704,
+    "upload_bytes_per_second": 5079199.718115679,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.10485184052848,
+    "estimated_full_shards_per_hour": 96.60799482662281,
     "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": 1.1277745370025514,
-      "bytes": 5019078,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.113636714999302,
       "bytes": 5851090,
@@ -562,6 +556,12 @@
       "bytes": 5980643,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.4380262719932944,
+      "bytes": 12216092,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -347,3 +347,4 @@
 {"step": 639100, "loss": 0.5460857152938843, "lr": 0.0001, "grad_norm": 0.1668767382361896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93548270150, "tokens_per_second": 151003.87434041852, "data_wait_s": 1.0772004316095263e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 639200, "loss": 0.575522243976593, "lr": 0.0001, "grad_norm": 0.2166116215513019, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93562733938, "tokens_per_second": 150187.81810608323, "data_wait_s": 1.72159998328425e-05, "gpu_peak_gb": 89.37861251831055}
 {"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}

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 sha256:85cd1b21830aafaef3892535ec4cd6bffea118a273ae7c139700341e3651850f
-size 744828
+oid sha256:05fe333866522386fe210a24a89804bd6b1243b432b8d6a03734a38ef71ab78d
+size 746946