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": 75844.1368570214,
-      "required_pages_per_second_upper_estimate": 0.005247073370411925
+      "required_token_positions_per_second": 76078.96140326493,
+      "required_pages_per_second_upper_estimate": 0.005263319077389068
     },
     {
       "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": 9480.517107127675,
-      "required_pages_per_second_upper_estimate": 0.003969294537044592
+      "required_token_positions_per_second": 9509.870175408116,
+      "required_pages_per_second_upper_estimate": 0.003981584053771845
     },
     {
       "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": 18961.03421425535,
-      "required_pages_per_second_upper_estimate": 0.003949083390731188
+      "required_token_positions_per_second": 19019.74035081623,
+      "required_pages_per_second_upper_estimate": 0.003961310330791888
     },
     {
       "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": 28441.551321383024,
-      "required_pages_per_second_upper_estimate": 0.001705027573982592
+      "required_token_positions_per_second": 28529.610526224347,
+      "required_pages_per_second_upper_estimate": 0.0017103065888542089
     },
     {
       "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": 28441.551321383024,
-      "required_pages_per_second_upper_estimate": 0.002211205514074366
+      "required_token_positions_per_second": 28529.610526224347,
+      "required_pages_per_second_upper_estimate": 0.0022180517299192713
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28441.551321383024,
-      "required_pages_per_second_upper_estimate": 0.3648176821921605
+      "required_token_positions_per_second": 28529.610526224347,
+      "required_pages_per_second_upper_estimate": 0.3659472111212574
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4029591803591188,
-    "producer_p95_seconds": 0.3505544009967707,
-    "gpu_update_seconds": 0.9720988734961721,
-    "producer_batches_per_second": 2.4816409421639087,
-    "gpu_batches_per_second": 1.0287019430477076,
-    "producer_headroom": 2.412400364299515,
+    "producer_batch_seconds": 0.15048974254659697,
+    "producer_p95_seconds": 0.3595450240027276,
+    "gpu_update_seconds": 0.9690984030288821,
+    "producer_batches_per_second": 6.644971165994018,
+    "gpu_batches_per_second": 1.0318869547968876,
+    "producer_headroom": 6.439630945137771,
     "sustainable": true,
-    "buffer_seconds": 60.27013015676267,
+    "buffer_seconds": 60.08410098779069,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324202.0926816801,
+    "generation_bytes_per_second": 332732.0886869041,
     "upload_bytes_per_second": 4184247.457970663,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.56622205102732,
+    "estimated_full_shards_per_hour": 71.39656062560408,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -393,3 +393,4 @@
 {"step": 839300, "loss": 0.546668291091919, "lr": 3e-05, "grad_norm": 0.3009325826473563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122849542127, "tokens_per_second": 149136.20504048473, "data_wait_s": 9.843999578151852e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 839400, "loss": 0.576280415058136, "lr": 3e-05, "grad_norm": 0.3093915645088455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122864220687, "tokens_per_second": 151340.57215926223, "data_wait_s": 1.3443997886497527e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 839500, "loss": 0.5308698415756226, "lr": 3e-05, "grad_norm": 0.28809007191134933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122878902505, "tokens_per_second": 151331.53464158325, "data_wait_s": 1.705899921944365e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 839600, "loss": 0.4512103497982025, "lr": 3e-05, "grad_norm": 0.30110378791240083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122893670748, "tokens_per_second": 152365.37923445462, "data_wait_s": 1.0973999451380223e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4882e9b3abad1044014bedde80c5aae4896ac16bbac4fdddffcecb6228cfb2d8
-size 842256
+oid sha256:ccb712bf5eee57b98aa4938a51d90e6b53ac7e053ed0eef7f1072c183aeb519e
+size 844374