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": 75786.18396517826,
-      "required_pages_per_second_upper_estimate": 0.006568954347774458
+      "required_token_positions_per_second": 75382.99396247821,
+      "required_pages_per_second_upper_estimate": 0.006534006860216136
     },
     {
       "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": 9473.272995647283,
-      "required_pages_per_second_upper_estimate": 0.003779030236016947
+      "required_token_positions_per_second": 9422.874245309777,
+      "required_pages_per_second_upper_estimate": 0.0037589254209788483
     },
     {
       "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": 18946.545991294566,
-      "required_pages_per_second_upper_estimate": 0.004382949967265209
+      "required_token_positions_per_second": 18845.748490619553,
+      "required_pages_per_second_upper_estimate": 0.0043596322394594674
     },
     {
       "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": 28419.81898694185,
-      "required_pages_per_second_upper_estimate": 0.001730617869038151
+      "required_token_positions_per_second": 28268.62273592933,
+      "required_pages_per_second_upper_estimate": 0.0017214108106169107
     },
     {
       "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": 28419.81898694185,
-      "required_pages_per_second_upper_estimate": 0.0021878363367089136
+      "required_token_positions_per_second": 28268.62273592933,
+      "required_pages_per_second_upper_estimate": 0.0021761968307679592
     },
     {
       "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": 28419.81898694185,
-      "required_pages_per_second_upper_estimate": 0.3528790368022381
+      "required_token_positions_per_second": 28268.62273592933,
+      "required_pages_per_second_upper_estimate": 0.3510016853859633
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.14072729456236743,
-    "producer_p95_seconds": 0.3794044540009054,
-    "gpu_update_seconds": 0.9728422272043155,
-    "producer_batches_per_second": 7.105942049904332,
-    "gpu_batches_per_second": 1.0279159066457555,
-    "producer_headroom": 6.912960490213729,
+    "producer_batch_seconds": 0.11223712345335457,
+    "producer_p95_seconds": 0.36896726800114266,
+    "gpu_update_seconds": 0.9780455262455882,
+    "producer_batches_per_second": 8.909708029140619,
+    "gpu_batches_per_second": 1.0224472922428143,
+    "producer_headroom": 8.714100078055377,
     "sustainable": true,
-    "buffer_seconds": 60.31621808666756,
+    "buffer_seconds": 60.63882262722647,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244730.08088644338,
-    "upload_bytes_per_second": 3764059.0242048367,
+    "generation_bytes_per_second": 226429.62254529845,
+    "upload_bytes_per_second": 3812385.173881963,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.51337833352066,
+    "estimated_full_shards_per_hour": 48.58652598637786,
     "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": 1.022158355001011,
-      "bytes": 1083307,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 33.35392404899903,
       "bytes": 16956987,
@@ -562,6 +556,12 @@
       "bytes": 12484307,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.181266768999194,
+      "bytes": 6999855,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -308,3 +308,4 @@
 {"step": 225800, "loss": 0.6041085869073868, "lr": 0.0003, "grad_norm": 0.1383085496420488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33050598731, "tokens_per_second": 146707.44813275008, "data_wait_s": 1.2013999366899952e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 225900, "loss": 0.5965267233550549, "lr": 0.0003, "grad_norm": 0.14059977196096476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33065426058, "tokens_per_second": 151548.53034548528, "data_wait_s": 1.4435001503443345e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 226000, "loss": 0.6034391447901726, "lr": 0.0003, "grad_norm": 0.16302936559749362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33080168587, "tokens_per_second": 149089.71686612893, "data_wait_s": 1.2119999155402184e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 226100, "loss": 0.6101761758327484, "lr": 0.0003, "grad_norm": 0.14038425550628747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33094931729, "tokens_per_second": 150510.12659799188, "data_wait_s": 1.3609998859465122e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:abe19bae3e247916be7dd21c5db79db4467701c5d328bc01ca0343563c48263b
-size 661443
+oid sha256:babceb02724dda0be48506b331f4327df5488c5d7027e01a4075bcaa1b32d960
+size 663561