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": 77189.14661511595,
-      "required_pages_per_second_upper_estimate": 0.005340124266329515
+      "required_token_positions_per_second": 77235.25295698272,
+      "required_pages_per_second_upper_estimate": 0.00534331401004131
     },
     {
       "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": 9648.643326889494,
-      "required_pages_per_second_upper_estimate": 0.0040396854743841625
+      "required_token_positions_per_second": 9654.40661962284,
+      "required_pages_per_second_upper_estimate": 0.004042098444700376
     },
     {
       "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": 19297.28665377899,
-      "required_pages_per_second_upper_estimate": 0.004019115906310814
+      "required_token_positions_per_second": 19308.81323924568,
+      "required_pages_per_second_upper_estimate": 0.0040215165900870245
     },
     {
       "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": 28945.92998066848,
-      "required_pages_per_second_upper_estimate": 0.001735264304465135
+      "required_token_positions_per_second": 28963.21985886852,
+      "required_pages_per_second_upper_estimate": 0.001736300806263112
     },
     {
       "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": 28945.92998066848,
-      "required_pages_per_second_upper_estimate": 0.002250418736306549
+      "required_token_positions_per_second": 28963.21985886852,
+      "required_pages_per_second_upper_estimate": 0.002251762948286465
     },
     {
       "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": 28945.92998066848,
-      "required_pages_per_second_upper_estimate": 0.37128731007386356
+      "required_token_positions_per_second": 28963.21985886852,
+      "required_pages_per_second_upper_estimate": 0.3715090860669889
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.28309694203130675,
-    "producer_p95_seconds": 0.25288008399729733,
-    "gpu_update_seconds": 0.9551601906888014,
-    "producer_batches_per_second": 3.532358890296361,
-    "gpu_batches_per_second": 1.0469448054350579,
-    "producer_headroom": 3.373968591236755,
+    "producer_batch_seconds": 0.08726010233601755,
+    "producer_p95_seconds": 0.2680390780005837,
+    "gpu_update_seconds": 0.9545899984436365,
+    "producer_batches_per_second": 11.459991144053921,
+    "gpu_batches_per_second": 1.0475701627194922,
+    "producer_headroom": 10.939592928366519,
     "sustainable": true,
-    "buffer_seconds": 60.17509201339449,
+    "buffer_seconds": 60.139169901949096,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340339.9147479333,
-    "upload_bytes_per_second": 5470724.6960292095,
+    "generation_bytes_per_second": 340755.37563627755,
+    "upload_bytes_per_second": 5585087.947976237,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.02902299717425,
+    "estimated_full_shards_per_hour": 73.11817123237843,
     "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": 0.95897385599892,
-      "bytes": 2279278,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.0730731740004558,
       "bytes": 6315193,
@@ -562,6 +556,12 @@
       "bytes": 7770394,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1173389949981356,
+      "bytes": 12086613,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -193,3 +193,4 @@
 {"step": 504300, "loss": 0.5694550275802612, "lr": 0.0003, "grad_norm": 0.16975562481172737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73812899236, "tokens_per_second": 151887.27632894603, "data_wait_s": 1.3411001418717206e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 504400, "loss": 0.5893999338150024, "lr": 0.0003, "grad_norm": 0.15345058368615985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73827340094, "tokens_per_second": 150455.18426398194, "data_wait_s": 1.47380014823284e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 504500, "loss": 0.5588216185569763, "lr": 0.0003, "grad_norm": 0.14769947251139195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73842026382, "tokens_per_second": 153087.75038712792, "data_wait_s": 1.0253999789711088e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 504600, "loss": 0.5975275635719299, "lr": 0.0003, "grad_norm": 0.1692339807413062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73856723021, "tokens_per_second": 153526.31281645794, "data_wait_s": 1.3040000339969993e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:871d29127beaf952b788f9968f8489251ad7ce59455f67401fe857f5518c764d
-size 415524
+oid sha256:23f037460c93e26580d9512f06c0d004e63c839ef6c13f677e3598718fdcae47
+size 417642