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": 76792.94619482545,
-      "required_pages_per_second_upper_estimate": 0.0053127142019421866
+      "required_token_positions_per_second": 77382.03158027382,
+      "required_pages_per_second_upper_estimate": 0.005353468495773916
     },
     {
       "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": 9599.11827435318,
-      "required_pages_per_second_upper_estimate": 0.004018950369087908
+      "required_token_positions_per_second": 9672.753947534227,
+      "required_pages_per_second_upper_estimate": 0.0040497800877611
     },
     {
       "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": 19198.23654870636,
-      "required_pages_per_second_upper_estimate": 0.003998486381551114
+      "required_token_positions_per_second": 19345.507895068455,
+      "required_pages_per_second_upper_estimate": 0.00402915911919176
     },
     {
       "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": 28797.35482305954,
-      "required_pages_per_second_upper_estimate": 0.001726357450627609
+      "required_token_positions_per_second": 29018.261842602682,
+      "required_pages_per_second_upper_estimate": 0.001739600489143732
     },
     {
       "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": 28797.35482305954,
-      "required_pages_per_second_upper_estimate": 0.002238867671315507
+      "required_token_positions_per_second": 29018.261842602682,
+      "required_pages_per_second_upper_estimate": 0.002256042220417696
     },
     {
       "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": 28797.35482305954,
-      "required_pages_per_second_upper_estimate": 0.36938154747963137
+      "required_token_positions_per_second": 29018.261842602682,
+      "required_pages_per_second_upper_estimate": 0.3722151055348531
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.086712057382897,
-    "producer_p95_seconds": 0.2482579799980158,
-    "gpu_update_seconds": 0.9600881806637603,
-    "producer_batches_per_second": 11.532421559141081,
-    "gpu_batches_per_second": 1.041570993310892,
-    "producer_headroom": 11.072141633363287,
+    "producer_batch_seconds": 0.12260932438286432,
+    "producer_p95_seconds": 0.2505970710008114,
+    "gpu_update_seconds": 0.9527793273754613,
+    "producer_batches_per_second": 8.155986545341069,
+    "gpu_batches_per_second": 1.0495609752098771,
+    "producer_headroom": 7.770855374753376,
     "sustainable": true,
-    "buffer_seconds": 60.4855553818169,
+    "buffer_seconds": 60.025097624654066,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334897.14684874646,
+    "generation_bytes_per_second": 320930.50915364356,
     "upload_bytes_per_second": 4748849.373259317,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.86113170716091,
+    "estimated_full_shards_per_hour": 68.86421638447743,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -219,3 +219,4 @@
 {"step": 781900, "loss": 0.4563373625278473, "lr": 3e-05, "grad_norm": 0.25067931439035296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114447799313, "tokens_per_second": 149144.83741606685, "data_wait_s": 1.950499790837057e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 782000, "loss": 0.5947047472000122, "lr": 3e-05, "grad_norm": 0.2786736856656454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114462427805, "tokens_per_second": 152177.5048435231, "data_wait_s": 1.8146998627344146e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 782100, "loss": 0.5047265291213989, "lr": 3e-05, "grad_norm": 0.2702417485551091, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114477135476, "tokens_per_second": 153693.59352186232, "data_wait_s": 1.4005003322381526e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 782200, "loss": 0.5140717029571533, "lr": 3e-05, "grad_norm": 0.2705502672653411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114491714452, "tokens_per_second": 135976.85138680084, "data_wait_s": 1.3278997357701883e-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 sha256:0a07742f83000016de5f00a3934d83630971e25f73967da605c205ec60f69207
-size 471375
+oid sha256:93b1126af115c25fdbc63e6eb53c9338715d05ae0290185b5263bf3d22406685
+size 473493