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": 77144.47345597814,
-      "required_pages_per_second_upper_estimate": 0.005337033673524351
+      "required_token_positions_per_second": 76937.63458860417,
+      "required_pages_per_second_upper_estimate": 0.005322724080747125
     },
     {
       "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": 9643.059181997267,
-      "required_pages_per_second_upper_estimate": 0.004037347509527993
+      "required_token_positions_per_second": 9617.20432357552,
+      "required_pages_per_second_upper_estimate": 0.004026522620217647
     },
     {
       "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": 19286.118363994534,
-      "required_pages_per_second_upper_estimate": 0.004016789846076319
+      "required_token_positions_per_second": 19234.40864715104,
+      "required_pages_per_second_upper_estimate": 0.004006020075734756
     },
     {
       "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": 28929.1775459918,
-      "required_pages_per_second_upper_estimate": 0.0017342600215857545
+      "required_token_positions_per_second": 28851.61297072656,
+      "required_pages_per_second_upper_estimate": 0.0017296101437328524
     },
     {
       "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": 28929.1775459918,
-      "required_pages_per_second_upper_estimate": 0.00224911630819661
+      "required_token_positions_per_second": 28851.61297072656,
+      "required_pages_per_second_upper_estimate": 0.0022430860036402483
     },
     {
       "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": 28929.1775459918,
-      "required_pages_per_second_upper_estimate": 0.3710724278291941
+      "required_token_positions_per_second": 28851.61297072656,
+      "required_pages_per_second_upper_estimate": 0.37007751274004386
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0747991208046841,
-    "producer_p95_seconds": 0.177951496998503,
-    "gpu_update_seconds": 0.9557133090301314,
-    "producer_batches_per_second": 13.369141097409498,
-    "gpu_batches_per_second": 1.0463388869354673,
-    "producer_headroom": 12.777066077095954,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.20993846889828,
+    "buffer_seconds": 60.371806656608435,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293530.60671955894,
+    "generation_bytes_per_second": 294101.6176916299,
     "upload_bytes_per_second": 4570984.797819422,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.984835159207115,
+    "estimated_full_shards_per_hour": 63.10736082135843,
     "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": [

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

@@ -88,3 +88,4 @@
 {"step": 449000, "loss": 0.5862565636634827, "lr": 0.0003, "grad_norm": 0.21102991124665044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65719775824, "tokens_per_second": 150423.05749013033, "data_wait_s": 1.5794999853824265e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:8e309934385eeba14fc991f80be30241e948842744cfee4e33c1699a83c18f41
-size 191568
+oid sha256:db6bca2f64f1ef811f34e5dd787f3e662c82c5a48272f18368a7122b90fc2462
+size 193686