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": 76937.63458860417,
-      "required_pages_per_second_upper_estimate": 0.005322724080747125
+      "required_token_positions_per_second": 77029.36085169048,
+      "required_pages_per_second_upper_estimate": 0.005329069916461689
     },
     {
       "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": 9617.20432357552,
-      "required_pages_per_second_upper_estimate": 0.004026522620217647
+      "required_token_positions_per_second": 9628.67010646131,
+      "required_pages_per_second_upper_estimate": 0.004031323104079153
     },
     {
       "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": 19234.40864715104,
-      "required_pages_per_second_upper_estimate": 0.004006020075734756
+      "required_token_positions_per_second": 19257.34021292262,
+      "required_pages_per_second_upper_estimate": 0.004010796116138919
     },
     {
       "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": 28851.61297072656,
-      "required_pages_per_second_upper_estimate": 0.0017296101437328524
+      "required_token_positions_per_second": 28886.01031938393,
+      "required_pages_per_second_upper_estimate": 0.0017316722122631514
     },
     {
       "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": 28851.61297072656,
-      "required_pages_per_second_upper_estimate": 0.0022430860036402483
+      "required_token_positions_per_second": 28886.01031938393,
+      "required_pages_per_second_upper_estimate": 0.0022457602462004115
     },
     {
       "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": 28851.61297072656,
-      "required_pages_per_second_upper_estimate": 0.37007751274004386
+      "required_token_positions_per_second": 28886.01031938393,
+      "required_pages_per_second_upper_estimate": 0.37051872499562516
     }
   ],
   "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": 0.07463661205476058,
-    "producer_p95_seconds": 0.184589317001155,
-    "gpu_update_seconds": 0.958282645342991,
-    "producer_batches_per_second": 13.398250167977936,
-    "gpu_batches_per_second": 1.0435334552490798,
-    "producer_headroom": 12.839310613937071,
+    "producer_batch_seconds": 0.05233814744531173,
+    "producer_p95_seconds": 0.18126035400018736,
+    "gpu_update_seconds": 0.9571415260987716,
+    "producer_batches_per_second": 19.106522657205293,
+    "gpu_batches_per_second": 1.0447775723156805,
+    "producer_headroom": 18.28764625455823,
     "sustainable": true,
-    "buffer_seconds": 60.371806656608435,
+    "buffer_seconds": 60.29991614422261,
     "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": 294101.6176916299,
-    "upload_bytes_per_second": 4570984.797819422,
+    "generation_bytes_per_second": 312779.1689010043,
+    "upload_bytes_per_second": 4574719.4528093655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.10736082135843,
+    "estimated_full_shards_per_hour": 67.11512851975057,
     "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": 1.1334343210000952,
-      "bytes": 6129536,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3353518999999778,
       "bytes": 13219093,
@@ -562,6 +556,12 @@
       "bytes": 2655368,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.023656208999455,
+      "bytes": 5985547,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -89,3 +89,4 @@
 {"step": 449100, "loss": 0.617741584777832, "lr": 0.0003, "grad_norm": 0.16030521457585725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65734252502, "tokens_per_second": 150393.0839166574, "data_wait_s": 1.53560013131937e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 449200, "loss": 0.6016988158226013, "lr": 0.0003, "grad_norm": 0.16539249928901117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65748872766, "tokens_per_second": 151996.4321911658, "data_wait_s": 1.2229000276420265e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 449300, "loss": 0.5931718349456787, "lr": 0.0003, "grad_norm": 0.14727832786485578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65763482923, "tokens_per_second": 152078.53892100567, "data_wait_s": 1.4055998690309934e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 449400, "loss": 0.5917835831642151, "lr": 0.0003, "grad_norm": 0.16826501700247157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65778230056, "tokens_per_second": 153351.57652372826, "data_wait_s": 1.408600110153202e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:db6bca2f64f1ef811f34e5dd787f3e662c82c5a48272f18368a7122b90fc2462
-size 193686
+oid sha256:73d35cc4a4a50c3d05839ae281b3618f965cc9e2eb05d81771eae71272440d51
+size 195804