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": 76837.68651538489,
-      "required_pages_per_second_upper_estimate": 0.005315809441130335
+      "required_token_positions_per_second": 77398.05627663089,
+      "required_pages_per_second_upper_estimate": 0.005354577121450316
     },
     {
       "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": 9604.710814423112,
-      "required_pages_per_second_upper_estimate": 0.004021291848829671
+      "required_token_positions_per_second": 9674.75703457886,
+      "required_pages_per_second_upper_estimate": 0.004050618738477473
     },
     {
       "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": 19209.421628846223,
-      "required_pages_per_second_upper_estimate": 0.004000815938773848
+      "required_token_positions_per_second": 19349.51406915772,
+      "required_pages_per_second_upper_estimate": 0.004029993499604639
     },
     {
       "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": 28814.132443269336,
-      "required_pages_per_second_upper_estimate": 0.0017273632433412434
+      "required_token_positions_per_second": 29024.271103736584,
+      "required_pages_per_second_upper_estimate": 0.001739960735173099
     },
     {
       "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": 28814.132443269336,
-      "required_pages_per_second_upper_estimate": 0.0022401720574898665
+      "required_token_positions_per_second": 29024.271103736584,
+      "required_pages_per_second_upper_estimate": 0.0022565094140389107
     },
     {
       "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": 28814.132443269336,
-      "required_pages_per_second_upper_estimate": 0.36959675277727755
+      "required_token_positions_per_second": 29024.271103736584,
+      "required_pages_per_second_upper_estimate": 0.3722921858844369
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11284612550767292,
-    "producer_p95_seconds": 0.3185121140013507,
-    "gpu_update_seconds": 0.959529149608607,
-    "producer_batches_per_second": 8.861624583929604,
-    "gpu_batches_per_second": 1.0421778227455634,
-    "producer_headroom": 8.5029871011687,
+    "producer_batch_seconds": 0.13272106581234766,
+    "producer_p95_seconds": 0.32685276700067334,
+    "gpu_update_seconds": 0.9525820614472071,
+    "producer_batches_per_second": 7.534598926547841,
+    "gpu_batches_per_second": 1.0497783240645466,
+    "producer_headroom": 7.177323777628856,
     "sustainable": true,
-    "buffer_seconds": 60.45033642534224,
+    "buffer_seconds": 60.01266987117405,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436091.238307728,
-    "upload_bytes_per_second": 4813585.146918953,
+    "generation_bytes_per_second": 435863.99902114295,
+    "upload_bytes_per_second": 4836393.792437452,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.57502805637246,
+    "estimated_full_shards_per_hour": 93.52626779533117,
     "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": 1.2271638120000716,
-      "bytes": 936260,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.1434297949999745,
       "bytes": 7632762,
@@ -562,6 +556,12 @@
       "bytes": 6121395,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.0959868669997377,
+      "bytes": 2315562,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -324,3 +324,4 @@
 {"step": 792400, "loss": 0.5232592225074768, "lr": 3e-05, "grad_norm": 0.25947454606977255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115984565712, "tokens_per_second": 151426.0776815659, "data_wait_s": 1.357600194751285e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 792500, "loss": 0.5563556551933289, "lr": 3e-05, "grad_norm": 0.26630299494470827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115999186454, "tokens_per_second": 152317.218110856, "data_wait_s": 1.4304998330771923e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 792600, "loss": 0.5497496128082275, "lr": 3e-05, "grad_norm": 0.2706153066848277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116013934420, "tokens_per_second": 153520.20475070903, "data_wait_s": 1.176499790744856e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 792700, "loss": 0.5422593355178833, "lr": 3e-05, "grad_norm": 0.26376113343857827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116028712499, "tokens_per_second": 152664.09741162587, "data_wait_s": 1.152500044554472e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f560b249ae54e438941d0d554939f41879cf2f7bbfe85654d8ff2c5c48c0893d
-size 695331
+oid sha256:413694cb9c5425415974f589a750e445ef6915c366c863d11104b72221b0c9a7
+size 697449