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": 76857.9917715777,
-      "required_pages_per_second_upper_estimate": 0.005317214205868438
+      "required_token_positions_per_second": 77211.70952706557,
+      "required_pages_per_second_upper_estimate": 0.005341685220931358
     },
     {
       "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": 9607.248971447212,
-      "required_pages_per_second_upper_estimate": 0.004022354522172917
+      "required_token_positions_per_second": 9651.463690883196,
+      "required_pages_per_second_upper_estimate": 0.004040866301892428
     },
     {
       "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": 19214.497942894424,
-      "required_pages_per_second_upper_estimate": 0.004001873201118685
+      "required_token_positions_per_second": 19302.927381766393,
+      "required_pages_per_second_upper_estimate": 0.0040202907211945894
     },
     {
       "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": 28821.746914341635,
-      "required_pages_per_second_upper_estimate": 0.0017278197192554037
+      "required_token_positions_per_second": 28954.39107264959,
+      "required_pages_per_second_upper_estimate": 0.001735771534010063
     },
     {
       "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": 28821.746914341635,
-      "required_pages_per_second_upper_estimate": 0.002240764049123219
+      "required_token_positions_per_second": 28954.39107264959,
+      "required_pages_per_second_upper_estimate": 0.0022510765489916697
     },
     {
       "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": 28821.746914341635,
-      "required_pages_per_second_upper_estimate": 0.369694423036411
+      "required_token_positions_per_second": 28954.39107264959,
+      "required_pages_per_second_upper_estimate": 0.3713958398769845
     }
   ],
   "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": 0.1044524166326255,
-    "producer_p95_seconds": 0.3575503110041609,
-    "gpu_update_seconds": 0.9592756498129688,
-    "producer_batches_per_second": 9.573737326893518,
-    "gpu_batches_per_second": 1.0424532304087686,
-    "producer_headroom": 9.183853095394452,
+    "producer_batch_seconds": 0.11941474192155965,
+    "producer_p95_seconds": 0.36903428599907784,
+    "gpu_update_seconds": 0.9548810724642174,
+    "producer_batches_per_second": 8.374175448596398,
+    "gpu_batches_per_second": 1.0472508345142357,
+    "producer_headroom": 7.996341633359248,
     "sustainable": true,
-    "buffer_seconds": 60.434365938217034,
+    "buffer_seconds": 60.157507565245695,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 173,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 539400.2541957784,
-    "upload_bytes_per_second": 5007926.911206305,
+    "generation_bytes_per_second": 733106.6774698836,
+    "upload_bytes_per_second": 4998683.824369287,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.7427379551412,
+    "estimated_full_shards_per_hour": 78.6538135675067,
     "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": 1.386075828995672,
-      "bytes": 6310823,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2331648629988194,
       "bytes": 6006485,
@@ -562,6 +556,12 @@
       "bytes": 7649400,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.5415065289998893,
+      "bytes": 6305077,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -421,3 +421,4 @@
 {"step": 967100, "loss": 0.5165047645568848, "lr": 3e-05, "grad_norm": 0.37873417304885043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141557278112, "tokens_per_second": 150595.3771794798, "data_wait_s": 1.3090997526887804e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 967200, "loss": 0.5619150996208191, "lr": 3e-05, "grad_norm": 0.3728332348666339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141571844518, "tokens_per_second": 151190.48189071094, "data_wait_s": 1.383299968438223e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 967300, "loss": 0.5057878494262695, "lr": 3e-05, "grad_norm": 0.35473595444178235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141586391668, "tokens_per_second": 151630.37521573255, "data_wait_s": 1.6806006897240877e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 967400, "loss": 0.5333903431892395, "lr": 3e-05, "grad_norm": 0.3574164218272315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141601135050, "tokens_per_second": 153021.78144090463, "data_wait_s": 1.2317999789956957e-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 sha256:6028886d130d85c09fe6a18d9e5f7c21032ec37ce283544c502dd2128c75bf84
-size 902343
+oid sha256:c35fac97da958f39deefa75fb9fb25ae8dd27b906c742f36003882961c641d22
+size 904461