Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75523.88110131283,
-      "required_pages_per_second_upper_estimate": 0.006546218600335133
+      "required_token_positions_per_second": 75506.02289831277,
+      "required_pages_per_second_upper_estimate": 0.006544670696560289
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9440.485137664104,
-      "required_pages_per_second_upper_estimate": 0.0037659506692452943
+      "required_token_positions_per_second": 9438.252862289097,
+      "required_pages_per_second_upper_estimate": 0.003765060181222713
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18880.970275328207,
-      "required_pages_per_second_upper_estimate": 0.004367780179469575
+      "required_token_positions_per_second": 18876.505724578194,
+      "required_pages_per_second_upper_estimate": 0.004366747384226969
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28321.455412992313,
-      "required_pages_per_second_upper_estimate": 0.0017246280434584124
+      "required_token_positions_per_second": 28314.75858686729,
+      "required_pages_per_second_upper_estimate": 0.0017242202418829298
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28321.455412992313,
-      "required_pages_per_second_upper_estimate": 0.002180264036498478
+      "required_token_positions_per_second": 28314.75858686729,
+      "required_pages_per_second_upper_estimate": 0.002179748496285373
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28321.455412992313,
-      "required_pages_per_second_upper_estimate": 0.35165769041548994
+      "required_token_positions_per_second": 28314.75858686729,
+      "required_pages_per_second_upper_estimate": 0.3515745382478524
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10977355503911213,
-    "producer_p95_seconds": 0.3447659509984078,
-    "gpu_update_seconds": 0.9762210167813845,
-    "producer_batches_per_second": 9.109662155367946,
-    "gpu_batches_per_second": 1.0243581963611224,
-    "producer_headroom": 8.893043651848195,
+    "producer_batch_seconds": 0.14284253294556493,
+    "producer_p95_seconds": 0.3418100549970404,
+    "gpu_update_seconds": 0.9764519063504733,
+    "producer_batches_per_second": 7.000715958887991,
+    "gpu_batches_per_second": 1.024115978981022,
+    "producer_headroom": 6.83586244387436,
     "sustainable": true,
-    "buffer_seconds": 60.52570304044584,
+    "buffer_seconds": 60.540018193729345,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328391.4827513423,
-    "upload_bytes_per_second": 6049124.401014321,
+    "generation_bytes_per_second": 316107.718365645,
+    "upload_bytes_per_second": 6110587.418275576,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.46516763596728,
+    "estimated_full_shards_per_hour": 67.82935775019656,
     "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": 1.171269268001197,
-      "bytes": 2613392,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.150122408002062,
       "bytes": 11429829,
@@ -562,6 +556,12 @@
       "bytes": 6920922,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9876665679985308,
+      "bytes": 5790107,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -261,3 +261,4 @@
 {"step": 111100, "loss": 0.643631000071764, "lr": 0.0003, "grad_norm": 0.14187004986412452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16257234626, "tokens_per_second": 148983.6677029596, "data_wait_s": 1.8320002709515393e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 111200, "loss": 0.6151546761393547, "lr": 0.0003, "grad_norm": 0.14512287742372176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16271827997, "tokens_per_second": 148151.61254821444, "data_wait_s": 7.84599978942424e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 111300, "loss": 0.6296935249119997, "lr": 0.0003, "grad_norm": 0.1266021879014762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16286494137, "tokens_per_second": 149323.95690671672, "data_wait_s": 1.8397993699181825e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 111400, "loss": 0.6208535227924585, "lr": 0.0003, "grad_norm": 0.142977013173407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16301263094, "tokens_per_second": 150214.72800092027, "data_wait_s": 1.4746001397725195e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d5b5e4900482c80afd0828520374e97a715dd7c53b9f1110fae42271a2f0f775
-size 561114
+oid sha256:4caa2c79f75a8b731b7f6b766daf8db5192b6e716d1d86d306b25b9268aa97ce
+size 563232