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": 77109.42094668193,
-      "required_pages_per_second_upper_estimate": 0.005334608659598212
+      "required_token_positions_per_second": 76521.99766604048,
+      "required_pages_per_second_upper_estimate": 0.005293969354033639
     },
     {
       "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": 9638.677618335241,
-      "required_pages_per_second_upper_estimate": 0.004035513040320156
+      "required_token_positions_per_second": 9565.24970825506,
+      "required_pages_per_second_upper_estimate": 0.004004770307718709
     },
     {
       "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": 19277.355236670483,
-      "required_pages_per_second_upper_estimate": 0.004014964717753928
+      "required_token_positions_per_second": 19130.49941651012,
+      "required_pages_per_second_upper_estimate": 0.003984378523262939
     },
     {
       "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": 28916.032855005724,
-      "required_pages_per_second_upper_estimate": 0.0017334720174319203
+      "required_token_positions_per_second": 28695.74912476518,
+      "required_pages_per_second_upper_estimate": 0.0017202663441577804
     },
     {
       "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": 28916.032855005724,
-      "required_pages_per_second_upper_estimate": 0.002248094366289828
+      "required_token_positions_per_second": 28695.74912476518,
+      "required_pages_per_second_upper_estimate": 0.002230968275189356
     },
     {
       "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": 28916.032855005724,
-      "required_pages_per_second_upper_estimate": 0.37090382184689424
+      "required_token_positions_per_second": 28695.74912476518,
+      "required_pages_per_second_upper_estimate": 0.36807825867761035
     }
   ],
   "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": 0.09417577299234381,
-    "producer_p95_seconds": 0.3254973750008503,
-    "gpu_update_seconds": 0.9561477585336811,
-    "producer_batches_per_second": 10.618442177069221,
-    "gpu_batches_per_second": 1.045863456850612,
-    "producer_headroom": 10.152799686724236,
+    "producer_batch_seconds": 0.3724495945313606,
+    "producer_p95_seconds": 0.3268765600005281,
+    "gpu_update_seconds": 0.9634876538608659,
+    "producer_batches_per_second": 2.6849270738454223,
+    "gpu_batches_per_second": 1.0378960186908703,
+    "producer_headroom": 2.5868940871668458,
     "sustainable": true,
-    "buffer_seconds": 60.237308787621906,
+    "buffer_seconds": 60.69972219323455,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290305.29844373395,
-    "upload_bytes_per_second": 4827192.109502443,
+    "generation_bytes_per_second": 272560.4078205887,
+    "upload_bytes_per_second": 4830505.51893794,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.29275908454908,
+    "estimated_full_shards_per_hour": 58.48511863673444,
     "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": 0.90897045499878,
-      "bytes": 2264477,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.6527253510030278,
       "bytes": 10383315,
@@ -562,6 +556,12 @@
       "bytes": 12830257,
       "pages": 24,
       "shards": 4
+    },
+    {
+      "seconds": 1.6963619910020498,
+      "bytes": 6365269,
+      "pages": 45,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -371,3 +371,4 @@
 {"step": 797100, "loss": 0.5762922167778015, "lr": 3e-05, "grad_norm": 0.2785145102406768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116672709376, "tokens_per_second": 152721.36811185308, "data_wait_s": 1.0238996765110642e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 797200, "loss": 0.5406336784362793, "lr": 3e-05, "grad_norm": 0.26895965673834704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116687364548, "tokens_per_second": 151971.33816354838, "data_wait_s": 1.3086995750200003e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 797300, "loss": 0.5212841629981995, "lr": 3e-05, "grad_norm": 0.27259592582662984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116702131641, "tokens_per_second": 153391.08720567572, "data_wait_s": 1.6286998288705945e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 797400, "loss": 0.5167383551597595, "lr": 3e-05, "grad_norm": 0.2813982906389843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116716638357, "tokens_per_second": 151536.85981189107, "data_wait_s": 1.4015000488143414e-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 sha256:86ca0de828b173da6c4a28802b72717cbfb1e70a937b9cfc55bdd4600649de32
-size 795660
+oid sha256:5bada24dab67e571579b3e92f3e3b60821c1bb408c968d9843848e67833370ea
+size 797778