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": 73932.32371296015,
-      "required_pages_per_second_upper_estimate": 0.006408266439678044
+      "required_token_positions_per_second": 74575.50378080607,
+      "required_pages_per_second_upper_estimate": 0.0064640156578339505
     },
     {
       "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": 9241.540464120018,
-      "required_pages_per_second_upper_estimate": 0.00368658866448062
+      "required_token_positions_per_second": 9321.93797260076,
+      "required_pages_per_second_upper_estimate": 0.0037186604326634592
     },
     {
       "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": 18483.080928240037,
-      "required_pages_per_second_upper_estimate": 0.004275735481634067
+      "required_token_positions_per_second": 18643.87594520152,
+      "required_pages_per_second_upper_estimate": 0.00431293257891246
     },
     {
       "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": 27724.621392360055,
-      "required_pages_per_second_upper_estimate": 0.001688283982948011
+      "required_token_positions_per_second": 27965.813917802276,
+      "required_pages_per_second_upper_estimate": 0.0017029713423080595
     },
     {
       "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": 27724.621392360055,
-      "required_pages_per_second_upper_estimate": 0.002134318101447896
+      "required_token_positions_per_second": 27965.813917802276,
+      "required_pages_per_second_upper_estimate": 0.0021528857697200813
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27724.621392360055,
-      "required_pages_per_second_upper_estimate": 0.34424700935421054
+      "required_token_positions_per_second": 27965.813917802276,
+      "required_pages_per_second_upper_estimate": 0.34724181330074716
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06659586385941196,
-    "producer_p95_seconds": 0.16490137100026914,
-    "gpu_update_seconds": 0.9972363412550993,
-    "producer_batches_per_second": 15.015947568621717,
-    "gpu_batches_per_second": 1.0027713177213562,
-    "producer_headroom": 14.974448613810726,
+    "producer_batch_seconds": 0.2908272009219033,
+    "producer_p95_seconds": 0.16686953800035553,
+    "gpu_update_seconds": 0.988635627815575,
+    "producer_batches_per_second": 3.4384679178222157,
+    "gpu_batches_per_second": 1.0114950057075476,
+    "producer_headroom": 3.3993918886598795,
     "sustainable": true,
-    "buffer_seconds": 60.83141681656106,
+    "buffer_seconds": 60.30677329675007,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 483939.5272402177,
-    "upload_bytes_per_second": 5847377.424965577,
+    "generation_bytes_per_second": 462046.69606518035,
+    "upload_bytes_per_second": 5853660.092391743,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.84215701012515,
+    "estimated_full_shards_per_hour": 99.14446507898863,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 2564532,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 0.9341291079999792,
+      "bytes": 5778135,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -54,3 +54,4 @@
 {"step": 87300, "loss": 0.6145474705845118, "lr": 0.0003, "grad_norm": 0.13754437614385157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12775735680, "tokens_per_second": 147150.29524200602, "data_wait_s": 1.011400036077248e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 87400, "loss": 0.6536234840750694, "lr": 0.0003, "grad_norm": 0.13182454596387277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12790268643, "tokens_per_second": 147874.8018019443, "data_wait_s": 9.637999937694985e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 87500, "loss": 0.6118768639862537, "lr": 0.0003, "grad_norm": 0.13208805657286993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12804833533, "tokens_per_second": 147901.13554139592, "data_wait_s": 9.451000551052857e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 87600, "loss": 0.6148227006196976, "lr": 0.0003, "grad_norm": 0.12887817111968602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12819477179, "tokens_per_second": 149061.14271705193, "data_wait_s": 1.0055000529973768e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b5eec56e3a12e1bbb21fc153d422872749e8a91c0f34aa702654fe245706faf9
-size 119556
+oid sha256:996c492c0961d2432120a260429487c87ec3d6824179a36c236d3e325d2ab2c0
+size 121674