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": 76048.61196042152,
-      "required_pages_per_second_upper_estimate": 0.00659170094128447
+      "required_token_positions_per_second": 75909.78872973534,
+      "required_pages_per_second_upper_estimate": 0.0065796680954928535
     },
     {
       "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": 9506.07649505269,
-      "required_pages_per_second_upper_estimate": 0.0037921160423857866
+      "required_token_positions_per_second": 9488.723591216918,
+      "required_pages_per_second_upper_estimate": 0.0037851937095966645
     },
     {
       "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": 19012.15299010538,
-      "required_pages_per_second_upper_estimate": 0.004398126991796347
+      "required_token_positions_per_second": 18977.447182433836,
+      "required_pages_per_second_upper_estimate": 0.004390098413992889
     },
     {
       "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": 28518.229485158066,
-      "required_pages_per_second_upper_estimate": 0.0017366105520595376
+      "required_token_positions_per_second": 28466.170773650752,
+      "required_pages_per_second_upper_estimate": 0.0017334404496596929
     },
     {
       "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": 28518.229485158066,
-      "required_pages_per_second_upper_estimate": 0.0021954122492792924
+      "required_token_positions_per_second": 28466.170773650752,
+      "required_pages_per_second_upper_estimate": 0.0021914046255597226
     },
     {
       "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": 28518.229485158066,
-      "required_pages_per_second_upper_estimate": 0.354100965831333
+      "required_token_positions_per_second": 28466.170773650752,
+      "required_pages_per_second_upper_estimate": 0.35345457086371174
     }
   ],
   "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": 0.07421308129676163,
-    "producer_p95_seconds": 0.27634474499791395,
-    "gpu_update_seconds": 0.9694851503452917,
-    "producer_batches_per_second": 13.474713386461103,
-    "gpu_batches_per_second": 1.0314753141333213,
-    "producer_headroom": 13.063534533332957,
+    "producer_batch_seconds": 0.08705815082035429,
+    "producer_p95_seconds": 0.29243760600002133,
+    "gpu_update_seconds": 0.9712581372409921,
+    "producer_batches_per_second": 11.486575243982772,
+    "gpu_batches_per_second": 1.0295924035608635,
+    "producer_headroom": 11.156429674749202,
     "sustainable": true,
-    "buffer_seconds": 60.108079321408084,
+    "buffer_seconds": 60.21800450894151,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295857.04368954815,
-    "upload_bytes_per_second": 5238317.9402646115,
+    "generation_bytes_per_second": 296200.5479070956,
+    "upload_bytes_per_second": 5199411.888868383,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.48403437628587,
+    "estimated_full_shards_per_hour": 63.55774238500262,
     "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": 1.1146210560000327,
-      "bytes": 5994268,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.7972013050002715,
       "bytes": 688709,
@@ -562,6 +556,12 @@
       "bytes": 17877863,
       "pages": 40,
       "shards": 5
+    },
+    {
+      "seconds": 0.7990716060012346,
+      "bytes": 1163166,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -177,3 +177,4 @@
 {"step": 172700, "loss": 0.5568112526088953, "lr": 0.0003, "grad_norm": 0.14181230128600136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25275678110, "tokens_per_second": 128728.2962534748, "data_wait_s": 1.0804000339703634e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 172800, "loss": 0.6099232845008373, "lr": 0.0003, "grad_norm": 0.14864510725580335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25290431838, "tokens_per_second": 150122.77375398081, "data_wait_s": 1.3106997357681394e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 172900, "loss": 0.6735049448907375, "lr": 0.0003, "grad_norm": 0.15338374036884048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25305230891, "tokens_per_second": 150855.46176527388, "data_wait_s": 1.589599924045615e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 173000, "loss": 0.605955695733428, "lr": 0.0003, "grad_norm": 0.14783632549958015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25319758593, "tokens_per_second": 148346.42206873198, "data_wait_s": 1.3387998478719965e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26316d036b4b21e77a4421b598c85476bb64582440551a8a5e5b8969a1d80693
-size 381636
+oid sha256:f2959e63099bf8f78c4b218aeebdad7b850ef644775e1cca75c4739f6debf050
+size 383754