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": 75753.33569931863,
-      "required_pages_per_second_upper_estimate": 0.0052407915356343345
+      "required_token_positions_per_second": 76368.21579001032,
+      "required_pages_per_second_upper_estimate": 0.00528333036702676
     },
     {
       "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": 9469.166962414829,
-      "required_pages_per_second_upper_estimate": 0.003964542468471297
+      "required_token_positions_per_second": 9546.02697375129,
+      "required_pages_per_second_upper_estimate": 0.003996722150198324
     },
     {
       "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": 18938.333924829658,
-      "required_pages_per_second_upper_estimate": 0.003944355519090706
+      "required_token_positions_per_second": 19092.05394750258,
+      "required_pages_per_second_upper_estimate": 0.003976371345943867
     },
     {
       "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": 28407.500887244485,
-      "required_pages_per_second_upper_estimate": 0.001702986302447989
+      "required_token_positions_per_second": 28638.08092125387,
+      "required_pages_per_second_upper_estimate": 0.0017168092234115255
     },
     {
       "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": 28407.500887244485,
-      "required_pages_per_second_upper_estimate": 0.002208558242592126
+      "required_token_positions_per_second": 28638.08092125387,
+      "required_pages_per_second_upper_estimate": 0.0022264848260219712
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28407.500887244485,
-      "required_pages_per_second_upper_estimate": 0.36438091978353904
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28638.08092125387,
+      "required_pages_per_second_upper_estimate": 0.36733855288225997
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.50974458599228,
-    "producer_p95_seconds": 0.2711016849971202,
-    "gpu_update_seconds": 0.9732640723920379,
-    "producer_batches_per_second": 1.9617667896430484,
-    "gpu_batches_per_second": 1.0274703735259145,
-    "producer_headroom": 1.9093171347714477,
+    "producer_batch_seconds": 0.8098722178517619,
+    "producer_p95_seconds": 0.27143339500253205,
+    "gpu_update_seconds": 0.965427818855031,
+    "producer_batches_per_second": 1.2347626921350188,
+    "gpu_batches_per_second": 1.0358102185059992,
+    "producer_headroom": 1.1920742526714774,
     "sustainable": true,
-    "buffer_seconds": 60.34237248830635,
+    "buffer_seconds": 60.82195258786695,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 30207.91724617346,
-    "upload_bytes_per_second": 4975898.004348678,
+    "generation_bytes_per_second": 30737.992186955864,
+    "upload_bytes_per_second": 4889308.79492777,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 6.481915836705235,
+    "estimated_full_shards_per_hour": 6.595657579484052,
     "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": 1.1388595060016087,
-      "bytes": 11609898,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.240830022998125,
       "bytes": 6131822,
@@ -562,6 +556,12 @@
       "bytes": 7063267,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0798259360017255,
+      "bytes": 4403386,
+      "pages": 18,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 711400, "loss": 0.5594947338104248, "lr": 8e-05, "grad_norm": 0.21541486985756092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104128815256, "tokens_per_second": 152396.34985180062, "data_wait_s": 1.97020017367322e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 711500, "loss": 0.5800469517707825, "lr": 8e-05, "grad_norm": 0.21757968886907136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104143551438, "tokens_per_second": 130549.11382430035, "data_wait_s": 1.709700154606253e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 711600, "loss": 0.5820253491401672, "lr": 8e-05, "grad_norm": 0.21575630043191904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104158248858, "tokens_per_second": 111436.34439939866, "data_wait_s": 1.4609002391807735e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 711700, "loss": 0.5247791409492493, "lr": 8e-05, "grad_norm": 0.21384825472213356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104172817531, "tokens_per_second": 151856.43157379143, "data_wait_s": 2.0504001440713182e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f491cf04ef22483d92a8e4ddd66fe459045ce5f5f0d3dd9b15f938e075b91dd7
-size 567468
+oid sha256:c4d371e66ac0befa46cec66edc533fd7e390c1b423c63d62f160c17d2451f1ba
+size 569586