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": 76063.97430091981,
-      "required_pages_per_second_upper_estimate": 0.005262282234873965
+      "required_token_positions_per_second": 75900.95195175639,
+      "required_pages_per_second_upper_estimate": 0.005251003970494874
     },
     {
       "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": 9507.996787614977,
-      "required_pages_per_second_upper_estimate": 0.003980799705423643
+      "required_token_positions_per_second": 9487.618993969549,
+      "required_pages_per_second_upper_estimate": 0.0039722679487610235
     },
     {
       "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": 19015.993575229953,
-      "required_pages_per_second_upper_estimate": 0.003960529976246383
+      "required_token_positions_per_second": 18975.237987939097,
+      "required_pages_per_second_upper_estimate": 0.003952041662210811
     },
     {
       "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": 28523.990362844932,
-      "required_pages_per_second_upper_estimate": 0.0017099696686411056
+      "required_token_positions_per_second": 28462.856981908648,
+      "required_pages_per_second_upper_estimate": 0.0017063048157992431
     },
     {
       "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": 28523.990362844932,
-      "required_pages_per_second_upper_estimate": 0.002217614786937008
+      "required_token_positions_per_second": 28462.856981908648,
+      "required_pages_per_second_upper_estimate": 0.0022128619354665363
     },
     {
       "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": 28523.990362844932,
-      "required_pages_per_second_upper_estimate": 0.3658751216998875
+      "required_token_positions_per_second": 28462.856981908648,
+      "required_pages_per_second_upper_estimate": 0.3650909683291472
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13058760665603586,
-    "producer_p95_seconds": 0.30750614799762843,
-    "gpu_update_seconds": 0.9692893472581596,
-    "producer_batches_per_second": 7.657694520996715,
-    "gpu_batches_per_second": 1.0316836792116946,
-    "producer_headroom": 7.422521723759291,
+    "producer_batch_seconds": 0.09131279939009573,
+    "producer_p95_seconds": 0.3108226879994618,
+    "gpu_update_seconds": 0.9713712160930794,
+    "producer_batches_per_second": 10.951367241824647,
+    "gpu_batches_per_second": 1.0294725470887098,
+    "producer_headroom": 10.63784291557312,
     "sustainable": true,
-    "buffer_seconds": 60.0959395300059,
+    "buffer_seconds": 60.225015397770925,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 476426.49547410815,
-    "upload_bytes_per_second": 5652778.768786177,
+    "generation_bytes_per_second": 476132.69551241276,
+    "upload_bytes_per_second": 5644051.338363622,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.23003528754647,
+    "estimated_full_shards_per_hour": 102.16699265507972,
     "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": 1.024204646000726,
-      "bytes": 6281125,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.1149510969989933,
       "bytes": 6170756,
@@ -562,6 +556,12 @@
       "bytes": 2714106,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.0593990999987,
+      "bytes": 5830929,
+      "pages": 27,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -367,3 +367,4 @@
 {"step": 721700, "loss": 0.5118348598480225, "lr": 8e-05, "grad_norm": 0.21285526642905558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105636903287, "tokens_per_second": 152914.39340622717, "data_wait_s": 1.3473996659740806e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721800, "loss": 0.603217601776123, "lr": 8e-05, "grad_norm": 0.21922386682163936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105651647851, "tokens_per_second": 153505.71169014205, "data_wait_s": 1.7553000361658633e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721900, "loss": 0.5191973447799683, "lr": 8e-05, "grad_norm": 0.24170040891430658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105666361451, "tokens_per_second": 153525.6795159405, "data_wait_s": 1.685399911366403e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 722000, "loss": 0.5707347393035889, "lr": 8e-05, "grad_norm": 0.2197074037151571, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105681078705, "tokens_per_second": 153069.99553446317, "data_wait_s": 1.587899896549061e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b3460f547a087c2cf24c4a08aee901e17648455a19e15fb493e806b459d8cd3
-size 787188
+oid sha256:ad01c7007454b2c57c716967d693138b85d56220a6fbf1a3133838af5e32cd1d
+size 789306