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": 76360.3439379551,
-      "required_pages_per_second_upper_estimate": 0.005282785774036366
+      "required_token_positions_per_second": 76872.21944973354,
+      "required_pages_per_second_upper_estimate": 0.005318198509656022
     },
     {
       "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": 9545.042992244387,
-      "required_pages_per_second_upper_estimate": 0.003996310177689254
+      "required_token_positions_per_second": 9609.027431216693,
+      "required_pages_per_second_upper_estimate": 0.004023099126139935
     },
     {
       "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": 19090.085984488775,
-      "required_pages_per_second_upper_estimate": 0.003975961471146766
+      "required_token_positions_per_second": 19218.054862433386,
+      "required_pages_per_second_upper_estimate": 0.004002614013656374
     },
     {
       "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": 28635.128976733162,
-      "required_pages_per_second_upper_estimate": 0.0017166322588448675
+      "required_token_positions_per_second": 28827.08229365008,
+      "required_pages_per_second_upper_estimate": 0.0017281395670982924
     },
     {
       "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": 28635.128976733162,
-      "required_pages_per_second_upper_estimate": 0.002226255325319718
+      "required_token_positions_per_second": 28827.08229365008,
+      "required_pages_per_second_upper_estimate": 0.0022411788513965972
     },
     {
       "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": 28635.128976733162,
-      "required_pages_per_second_upper_estimate": 0.3673006885074994
+      "required_token_positions_per_second": 28827.08229365008,
+      "required_pages_per_second_upper_estimate": 0.36976285955349575
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11463898049231602,
-    "producer_p95_seconds": 0.34030379300384084,
-    "gpu_update_seconds": 0.9655273430919333,
-    "producer_batches_per_second": 8.72303640267481,
-    "gpu_batches_per_second": 1.0357034496792954,
-    "producer_headroom": 8.422330161568825,
+    "producer_batch_seconds": 0.11602489262469362,
+    "producer_p95_seconds": 0.35435975700238487,
+    "gpu_update_seconds": 0.9590981049820014,
+    "producer_batches_per_second": 8.618840124547287,
+    "gpu_batches_per_second": 1.0426462056441725,
+    "producer_headroom": 8.26631323059614,
     "sustainable": true,
-    "buffer_seconds": 60.8282226147918,
+    "buffer_seconds": 60.42318061386609,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344964.2551620706,
-    "upload_bytes_per_second": 5068410.633887698,
+    "generation_bytes_per_second": 329669.03623812355,
+    "upload_bytes_per_second": 5119148.843034814,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.02129880091275,
+    "estimated_full_shards_per_hour": 70.73930087430745,
     "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": 1.211611481005093,
-      "bytes": 2609258,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0486429719967418,
       "bytes": 6039932,
@@ -562,6 +556,12 @@
       "bytes": 1277647,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.0343618879996939,
+      "bytes": 5958933,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -417,3 +417,4 @@
 {"step": 438800, "loss": 0.582893967628479, "lr": 0.0003, "grad_norm": 0.1656231395262307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64227561788, "tokens_per_second": 151095.56874685982, "data_wait_s": 1.3587996363639832e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 438900, "loss": 0.5672533512115479, "lr": 0.0003, "grad_norm": 0.15173435938648816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64242189125, "tokens_per_second": 151674.12938917405, "data_wait_s": 1.3265998859424144e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 439000, "loss": 0.6220097541809082, "lr": 0.0003, "grad_norm": 0.18819345086694877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64256836105, "tokens_per_second": 151936.96073377918, "data_wait_s": 1.0144998668693006e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 439100, "loss": 0.5693238973617554, "lr": 0.0003, "grad_norm": 0.1412137916417348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64271512797, "tokens_per_second": 152294.6378369886, "data_wait_s": 8.383998647332191e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e2c36f9fafc29502a595530178966b2004ea99d3cbb207353859b8ea7d3e1525
-size 893871
+oid sha256:f253c0f13e5cbab9b9d55e76170d4c5aaee36c1a3416cd4a32b6edba2c041096
+size 895989