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": 77191.56523177953,
-      "required_pages_per_second_upper_estimate": 0.005340291591842265
+      "required_token_positions_per_second": 77201.88168552436,
+      "required_pages_per_second_upper_estimate": 0.005341005308049803
     },
     {
       "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": 9648.945653972442,
-      "required_pages_per_second_upper_estimate": 0.004039812052420485
+      "required_token_positions_per_second": 9650.235210690546,
+      "required_pages_per_second_upper_estimate": 0.004040351962889349
     },
     {
       "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": 19297.891307944883,
-      "required_pages_per_second_upper_estimate": 0.004019241839827767
+      "required_token_positions_per_second": 19300.47042138109,
+      "required_pages_per_second_upper_estimate": 0.004019779001140734
     },
     {
       "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": 28946.836961917325,
-      "required_pages_per_second_upper_estimate": 0.001735318676606123
+      "required_token_positions_per_second": 28950.705632071637,
+      "required_pages_per_second_upper_estimate": 0.0017355505974747567
     },
     {
       "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": 28946.836961917325,
-      "required_pages_per_second_upper_estimate": 0.002250489250109258
+      "required_token_positions_per_second": 28950.705632071637,
+      "required_pages_per_second_upper_estimate": 0.002250790022197274
     },
     {
       "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": 28946.836961917325,
-      "required_pages_per_second_upper_estimate": 0.3712989438554832
+      "required_token_positions_per_second": 28950.705632071637,
+      "required_pages_per_second_upper_estimate": 0.37134856700236835
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.14534221102383071,
-    "producer_p95_seconds": 0.36405883300176356,
-    "gpu_update_seconds": 0.9551302629843087,
-    "producer_batches_per_second": 6.8803136608128055,
-    "gpu_batches_per_second": 1.0469776100230512,
-    "producer_headroom": 6.571595796266666,
+    "producer_batch_seconds": 0.09866105283589377,
+    "producer_p95_seconds": 0.3659463359945221,
+    "gpu_update_seconds": 0.955002629344257,
+    "producer_batches_per_second": 10.13571182605697,
+    "gpu_batches_per_second": 1.0471175358822207,
+    "producer_headroom": 9.679631444160085,
     "sustainable": true,
-    "buffer_seconds": 60.173206568011445,
+    "buffer_seconds": 60.16516564868819,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 623478.6374832251,
-    "upload_bytes_per_second": 5042351.457843713,
+    "generation_bytes_per_second": 623486.9449359304,
+    "upload_bytes_per_second": 5011015.678262131,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.89200088201792,
+    "estimated_full_shards_per_hour": 66.8928921749994,
     "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": 1.5529981709987624,
-      "bytes": 11067080,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.1875661560043227,
       "bytes": 6040644,
@@ -562,6 +556,12 @@
       "bytes": 2703084,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0808033600042108,
+      "bytes": 6031307,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -430,3 +430,4 @@
 {"step": 968000, "loss": 0.5415030121803284, "lr": 3e-05, "grad_norm": 0.33291041638664554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141688819063, "tokens_per_second": 151903.86568822814, "data_wait_s": 9.25700442166999e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 968100, "loss": 0.5158685445785522, "lr": 3e-05, "grad_norm": 0.3647784515903567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141703541544, "tokens_per_second": 152873.7490072066, "data_wait_s": 1.3835997378919274e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 968200, "loss": 0.6089076995849609, "lr": 3e-05, "grad_norm": 0.4069208616623038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141718161237, "tokens_per_second": 152086.39543300463, "data_wait_s": 1.3898999895900488e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 968300, "loss": 0.4854298233985901, "lr": 3e-05, "grad_norm": 0.3531872913909241, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141732946685, "tokens_per_second": 153001.29599698883, "data_wait_s": 1.1126001481898129e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:28fb34cc5f3fd1c587dbe74ad7732a26ccbfd4235867a0f1ae13d3963ebc3fb0
-size 921405
+oid sha256:12f98381bb5f083b4ca76880be92eb721a2a5dc9d2e4ffd9ae65100d8b497235
+size 923523