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": 77264.84393711996,
-      "required_pages_per_second_upper_estimate": 0.0053453611827077386
+      "required_token_positions_per_second": 75327.77583149084,
+      "required_pages_per_second_upper_estimate": 0.005211350316543073
     },
     {
       "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": 9658.105492139995,
-      "required_pages_per_second_upper_estimate": 0.004043647085382068
+      "required_token_positions_per_second": 9415.971978936355,
+      "required_pages_per_second_upper_estimate": 0.003942270839726433
     },
     {
       "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": 19316.21098427999,
-      "required_pages_per_second_upper_estimate": 0.004023057345286016
+      "required_token_positions_per_second": 18831.94395787271,
+      "required_pages_per_second_upper_estimate": 0.003922197294665677
     },
     {
       "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": 28974.316476419986,
-      "required_pages_per_second_upper_estimate": 0.0017369660315417685
+      "required_token_positions_per_second": 28247.915936809066,
+      "required_pages_per_second_upper_estimate": 0.0016934194283414958
     },
     {
       "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": 28974.316476419986,
-      "required_pages_per_second_upper_estimate": 0.0022526256614922293
+      "required_token_positions_per_second": 28247.915936809066,
+      "required_pages_per_second_upper_estimate": 0.0021961512146357856
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28974.316476419986,
-      "required_pages_per_second_upper_estimate": 0.37165142156231945
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28247.915936809066,
+      "required_pages_per_second_upper_estimate": 0.3623339353883232
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.33411507805493557,
-    "producer_p95_seconds": 0.3525845330004813,
-    "gpu_update_seconds": 0.9542244084515549,
-    "producer_batches_per_second": 2.992980759268754,
-    "gpu_batches_per_second": 1.0479715160742182,
-    "producer_headroom": 2.8559752945201122,
+    "producer_batch_seconds": 0.15345990973452217,
+    "producer_p95_seconds": 0.37692647999938345,
+    "gpu_update_seconds": 0.9787624708969297,
+    "producer_batches_per_second": 6.516359886630646,
+    "gpu_batches_per_second": 1.0216983484088926,
+    "producer_headroom": 6.377968503892247,
     "sustainable": true,
-    "buffer_seconds": 60.11613773244796,
+    "buffer_seconds": 60.683273195609644,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 510515.22026218585,
-    "upload_bytes_per_second": 4477197.738613016,
+    "generation_bytes_per_second": 496398.35036077356,
+    "upload_bytes_per_second": 4617267.088236306,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.5446820821684,
+    "estimated_full_shards_per_hour": 106.5155304252377,
     "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": 1.2645733180033858,
-      "bytes": 2430189,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.1291347480000695,
       "bytes": 6037077,
@@ -562,6 +556,12 @@
       "bytes": 6369388,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.4435183870009496,
+      "bytes": 16576979,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -430,3 +430,4 @@
 {"step": 803000, "loss": 0.5340162515640259, "lr": 3e-05, "grad_norm": 0.274242308446818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117535845279, "tokens_per_second": 151410.89411223173, "data_wait_s": 1.251199864782393e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 803100, "loss": 0.5561849474906921, "lr": 3e-05, "grad_norm": 0.2756510483160743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117550351913, "tokens_per_second": 151219.05238726872, "data_wait_s": 1.7641999875195324e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 803200, "loss": 0.5094794631004333, "lr": 3e-05, "grad_norm": 0.26051532754282786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117565077044, "tokens_per_second": 150728.77412666593, "data_wait_s": 9.891002264339477e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 803300, "loss": 0.5157017707824707, "lr": 3e-05, "grad_norm": 0.2845687018671663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117579791596, "tokens_per_second": 152538.1604435676, "data_wait_s": 1.0247997124679387e-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:c568b3a35c265a22b5f5b978c179b1379e4c30d29bee0bdb3b3aaec505b1f394
-size 921405
+oid sha256:151cd75a5719dc073972e2ace5a20984c108512a055275c4a521429f2f778e53
+size 923523