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": 75783.37983466947,
-      "required_pages_per_second_upper_estimate": 0.006568711292848959
+      "required_token_positions_per_second": 75236.0254085183,
+      "required_pages_per_second_upper_estimate": 0.006521267998447286
     },
     {
       "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": 9472.922479333683,
-      "required_pages_per_second_upper_estimate": 0.003778890409818766
+      "required_token_positions_per_second": 9404.503176064787,
+      "required_pages_per_second_upper_estimate": 0.0037515969267850596
     },
     {
       "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": 18945.844958667367,
-      "required_pages_per_second_upper_estimate": 0.004382787795704658
+      "required_token_positions_per_second": 18809.006352129574,
+      "required_pages_per_second_upper_estimate": 0.004351132592359363
     },
     {
       "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": 28418.767438001047,
-      "required_pages_per_second_upper_estimate": 0.001730553835224707
+      "required_token_positions_per_second": 28213.509528194358,
+      "required_pages_per_second_upper_estimate": 0.0017180547054225048
     },
     {
       "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": 28418.767438001047,
-      "required_pages_per_second_upper_estimate": 0.0021877553855605765
+      "required_token_positions_per_second": 28213.509528194358,
+      "required_pages_per_second_upper_estimate": 0.002171954063473398
     },
     {
       "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": 28418.767438001047,
-      "required_pages_per_second_upper_estimate": 0.35286598008370124
+      "required_token_positions_per_second": 28213.509528194358,
+      "required_pages_per_second_upper_estimate": 0.3503173637979358
     }
   ],
   "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": 0.15381765184378082,
-    "producer_p95_seconds": 0.43848417999834055,
-    "gpu_update_seconds": 0.9728782242339479,
-    "producer_batches_per_second": 6.501204432737101,
-    "gpu_batches_per_second": 1.0278778731915887,
-    "producer_headroom": 6.324880223903142,
+    "producer_batch_seconds": 0.1356778864300736,
+    "producer_p95_seconds": 0.45626032799918903,
+    "gpu_update_seconds": 0.979956072900848,
+    "producer_batches_per_second": 7.370397832040119,
+    "gpu_batches_per_second": 1.020453903652863,
+    "producer_headroom": 7.2226661152029585,
     "sustainable": true,
-    "buffer_seconds": 60.31844990250477,
+    "buffer_seconds": 60.757276519852574,
     "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": 265967.61739024066,
-    "upload_bytes_per_second": 6163943.471077861,
+    "generation_bytes_per_second": 277686.27140744106,
+    "upload_bytes_per_second": 6217501.474770181,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.07045928268828,
+    "estimated_full_shards_per_hour": 59.58500963847565,
     "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": 1.576678247001837,
-      "bytes": 2288077,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.2293803360007587,
       "bytes": 10631332,
@@ -562,6 +556,12 @@
       "bytes": 5836559,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.0610048369999276,
+      "bytes": 2712570,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -413,3 +413,4 @@
 {"step": 81300, "loss": 0.6381821688264608, "lr": 0.0003, "grad_norm": 0.13018978879187185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11897616997, "tokens_per_second": 149945.90270480458, "data_wait_s": 9.181007044389844e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 81400, "loss": 0.5682158749550581, "lr": 0.0003, "grad_norm": 0.14436383490892077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11912344288, "tokens_per_second": 149654.9083027166, "data_wait_s": 1.3185002899263054e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 81500, "loss": 0.6229772474616766, "lr": 0.0003, "grad_norm": 0.13445697579748267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11927072858, "tokens_per_second": 148503.2866475685, "data_wait_s": 1.4524994185194373e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 81600, "loss": 0.6269095726311207, "lr": 0.0003, "grad_norm": 0.13646336345421883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11941723632, "tokens_per_second": 148375.01129323352, "data_wait_s": 8.81299638422206e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ac0f2b6f6ede3252b93ea32be157304fd12db66669c0e4d322089b5ead70680
-size 885399
+oid sha256:5e25bc04af67f1aea61e3fb6b2012453415d35493f1161d422c33aea6e53b3df
+size 887517