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": 77080.5736905296,
-      "required_pages_per_second_upper_estimate": 0.005332612939482738
+      "required_token_positions_per_second": 76353.78779250836,
+      "required_pages_per_second_upper_estimate": 0.005282332204682007
     },
     {
       "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": 9635.0717113162,
-      "required_pages_per_second_upper_estimate": 0.004034003322351185
+      "required_token_positions_per_second": 9544.223474063545,
+      "required_pages_per_second_upper_estimate": 0.003995967062540421
     },
     {
       "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": 19270.1434226324,
-      "required_pages_per_second_upper_estimate": 0.004013462687078122
+      "required_token_positions_per_second": 19088.44694812709,
+      "required_pages_per_second_upper_estimate": 0.0039756201030969225
     },
     {
       "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": 28905.2151339486,
-      "required_pages_per_second_upper_estimate": 0.0017328235115722495
+      "required_token_positions_per_second": 28632.670422190637,
+      "required_pages_per_second_upper_estimate": 0.0017164848722540384
     },
     {
       "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": 28905.2151339486,
-      "required_pages_per_second_upper_estimate": 0.0022472533360597628
+      "required_token_positions_per_second": 28632.670422190637,
+      "required_pages_per_second_upper_estimate": 0.002226064183518075
     },
     {
       "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": 28905.2151339486,
-      "required_pages_per_second_upper_estimate": 0.3707650637363374
+      "required_token_positions_per_second": 28632.670422190637,
+      "required_pages_per_second_upper_estimate": 0.3672691528096181
     }
   ],
   "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": 0.11596354778100704,
-    "producer_p95_seconds": 0.33427969200420193,
-    "gpu_update_seconds": 0.9565055949896035,
-    "producer_batches_per_second": 8.623399500405625,
-    "gpu_batches_per_second": 1.0454721908980253,
-    "producer_headroom": 8.248329869968533,
+    "producer_batch_seconds": 0.15320810067993307,
+    "producer_p95_seconds": 0.35817955300444737,
+    "gpu_update_seconds": 0.9656102484444655,
+    "producer_batches_per_second": 6.527070014979817,
+    "gpu_batches_per_second": 1.0356145262655756,
+    "producer_headroom": 6.302605698779082,
     "sustainable": true,
-    "buffer_seconds": 60.259852484345025,
+    "buffer_seconds": 60.83344565200133,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345247.3613093276,
-    "upload_bytes_per_second": 4972289.798497308,
+    "generation_bytes_per_second": 389733.43275672355,
+    "upload_bytes_per_second": 4967457.460324604,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.08204678974029,
+    "estimated_full_shards_per_hour": 83.62772213960915,
     "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": 1.0347107170018717,
-      "bytes": 6246238,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.040108234003128,
       "bytes": 901723,
@@ -562,6 +556,12 @@
       "bytes": 9557603,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.387810526000976,
+      "bytes": 7581888,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -436,3 +436,4 @@
 {"step": 928600, "loss": 0.5859284996986389, "lr": 3e-05, "grad_norm": 0.3761148145493249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135920801717, "tokens_per_second": 153091.6034122629, "data_wait_s": 2.195099659729749e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 928700, "loss": 0.5363896489143372, "lr": 3e-05, "grad_norm": 0.3356433281879512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135935459183, "tokens_per_second": 153489.3414576001, "data_wait_s": 1.1211996024940163e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 928800, "loss": 0.5777209997177124, "lr": 3e-05, "grad_norm": 0.3457022501605244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135949889323, "tokens_per_second": 150793.16417638108, "data_wait_s": 1.3526005204766989e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 928900, "loss": 0.5195816159248352, "lr": 3e-05, "grad_norm": 0.3415889347990047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135964381838, "tokens_per_second": 151530.63202996302, "data_wait_s": 1.6445999790448695e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:910618203ad29b0759d031a310e45abdea8c259ab0d43fce8ca9fdb99549ce2e
-size 934113
+oid sha256:f21d354867d8df7ebbe236ef88f64662be43a85e1a7a1323933661e1e381e6ce
+size 936231