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": 77176.33676402176,
-      "required_pages_per_second_upper_estimate": 0.005339238051107869
+      "required_token_positions_per_second": 77015.96986521692,
+      "required_pages_per_second_upper_estimate": 0.005328143496946079
     },
     {
       "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": 9647.04209550272,
-      "required_pages_per_second_upper_estimate": 0.004039015072240034
+      "required_token_positions_per_second": 9626.996233152115,
+      "required_pages_per_second_upper_estimate": 0.004030622288279042
     },
     {
       "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": 19294.08419100544,
-      "required_pages_per_second_upper_estimate": 0.004018448917769706
+      "required_token_positions_per_second": 19253.99246630423,
+      "required_pages_per_second_upper_estimate": 0.004010098868804324
     },
     {
       "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": 28941.126286508163,
-      "required_pages_per_second_upper_estimate": 0.0017349763303350402
+      "required_token_positions_per_second": 28880.988699456342,
+      "required_pages_per_second_upper_estimate": 0.0017313711738160622
     },
     {
       "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": 28941.126286508163,
-      "required_pages_per_second_upper_estimate": 0.002250045270214802
+      "required_token_positions_per_second": 28880.988699456342,
+      "required_pages_per_second_upper_estimate": 0.0022453698373388134
     },
     {
       "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": 28941.126286508163,
-      "required_pages_per_second_upper_estimate": 0.37122569344298
+      "required_token_positions_per_second": 28880.988699456342,
+      "required_pages_per_second_upper_estimate": 0.37045431304698945
     }
   ],
   "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": 0.12006736707803611,
-    "producer_p95_seconds": 0.31453827299992554,
-    "gpu_update_seconds": 0.9553187296960521,
-    "producer_batches_per_second": 8.328657688896135,
-    "gpu_batches_per_second": 1.0467710607099305,
-    "producer_headroom": 7.956522683429512,
+    "producer_batch_seconds": 0.13360909560176992,
+    "producer_p95_seconds": 0.3216207929981465,
+    "gpu_update_seconds": 0.9573079470274661,
+    "producer_batches_per_second": 7.48452038759817,
+    "gpu_batches_per_second": 1.0445959454375124,
+    "producer_headroom": 7.16499084673682,
     "sustainable": true,
-    "buffer_seconds": 60.185079970851284,
+    "buffer_seconds": 60.310400662730366,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382589.9067359196,
-    "upload_bytes_per_second": 4904583.399781482,
+    "generation_bytes_per_second": 437490.5467860291,
+    "upload_bytes_per_second": 4810305.035074043,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.09488774831954,
+    "estimated_full_shards_per_hour": 93.87528708158165,
     "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": 1.4039977110005566,
-      "bytes": 12182760,
-      "pages": 48,
-      "shards": 3
-    },
     {
       "seconds": 1.015954297999997,
       "bytes": 2619113,
@@ -562,6 +556,12 @@
       "bytes": 6035033,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1105879839997215,
+      "bytes": 2480881,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -309,3 +309,4 @@
 {"step": 955900, "loss": 0.5105063319206238, "lr": 3e-05, "grad_norm": 0.3532509704132093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139917613706, "tokens_per_second": 152705.56575056832, "data_wait_s": 1.0668001777958125e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 956000, "loss": 0.5017450451850891, "lr": 3e-05, "grad_norm": 0.3368989881529334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139932144721, "tokens_per_second": 151141.53548767915, "data_wait_s": 1.2212000001454726e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 956100, "loss": 0.5179443955421448, "lr": 3e-05, "grad_norm": 0.35472210072890226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139946608043, "tokens_per_second": 150216.62423743444, "data_wait_s": 1.1257001460762694e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 956200, "loss": 0.4910849332809448, "lr": 3e-05, "grad_norm": 0.3504413896596116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139961123319, "tokens_per_second": 150464.80419878374, "data_wait_s": 1.5675999748054892e-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:aeb6b0e35c6d22d682206063b917d471b5b392d8cb50c9d8d82f02af231e76f7
-size 663561
+oid sha256:f271c16ba6112055abae4da94ba0cb20c47ce17985c7774ac3e2446da834e443
+size 665679