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": 74431.48136062379,
-      "required_pages_per_second_upper_estimate": 0.006451532159474037
+      "required_token_positions_per_second": 75580.75028977283,
+      "required_pages_per_second_upper_estimate": 0.006551147877457204
     },
     {
       "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": 9303.935170077973,
-      "required_pages_per_second_upper_estimate": 0.0037114788455712357
+      "required_token_positions_per_second": 9447.593786221603,
+      "required_pages_per_second_upper_estimate": 0.003768786415438648
     },
     {
       "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": 18607.870340155947,
-      "required_pages_per_second_upper_estimate": 0.004304603315862173
+      "required_token_positions_per_second": 18895.187572443207,
+      "required_pages_per_second_upper_estimate": 0.004371069100941247
     },
     {
       "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": 27911.80551023392,
-      "required_pages_per_second_upper_estimate": 0.001699682513647362
+      "required_token_positions_per_second": 28342.78135866481,
+      "required_pages_per_second_upper_estimate": 0.0017259266816612778
     },
     {
       "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": 27911.80551023392,
-      "required_pages_per_second_upper_estimate": 0.002148728053000628
+      "required_token_positions_per_second": 28342.78135866481,
+      "required_pages_per_second_upper_estimate": 0.002181905766830343
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27911.80551023392,
-      "required_pages_per_second_upper_estimate": 0.34657120963326077
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28342.78135866481,
+      "required_pages_per_second_upper_estimate": 0.3519224872873935
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4876481998437612,
-    "producer_p95_seconds": 0.31666012599816895,
-    "gpu_update_seconds": 0.9905486045989682,
-    "producer_batches_per_second": 2.050658651709147,
-    "gpu_batches_per_second": 1.0095415766143634,
-    "producer_headroom": 2.0312770659592974,
+    "producer_batch_seconds": 0.1161214933909207,
+    "producer_p95_seconds": 0.32137131999843405,
+    "gpu_update_seconds": 0.9754864792600036,
+    "producer_batches_per_second": 8.611670163709658,
+    "gpu_batches_per_second": 1.0251295340952262,
+    "producer_headroom": 8.400567808545553,
     "sustainable": true,
-    "buffer_seconds": 60.42346488053706,
+    "buffer_seconds": 60.48016171412023,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313131.51838341577,
-    "upload_bytes_per_second": 5727897.552657762,
+    "generation_bytes_per_second": 328650.2283092073,
+    "upload_bytes_per_second": 5790414.661888446,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.1907345163999,
+    "estimated_full_shards_per_hour": 70.52068840939678,
     "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": 1.0752307399998244,
-      "bytes": 5923485,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.4506824640011473,
       "bytes": 7595330,
@@ -562,6 +556,12 @@
       "bytes": 6075598,
       "pages": 20,
       "shards": 2
+    },
+    {
+      "seconds": 1.3699396060001163,
+      "bytes": 12184416,
+      "pages": 47,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -181,3 +181,4 @@
 {"step": 133100, "loss": 0.6408717334270477, "lr": 0.0003, "grad_norm": 0.15209618771773498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19478356154, "tokens_per_second": 150286.3180154748, "data_wait_s": 1.3798999134451151e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 133200, "loss": 0.629723247140646, "lr": 0.0003, "grad_norm": 0.15105344666535644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19493198064, "tokens_per_second": 129447.22662660244, "data_wait_s": 1.0488998668733984e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 133300, "loss": 0.6020955741405487, "lr": 0.0003, "grad_norm": 0.13759156328564195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19507963850, "tokens_per_second": 150550.9122177826, "data_wait_s": 1.0233001376036555e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 133400, "loss": 0.5951715297996998, "lr": 0.0003, "grad_norm": 0.1380262139766603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19522758738, "tokens_per_second": 150925.6095559784, "data_wait_s": 1.2779000826412812e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba6a418e0463cc60614897e3431e80cb2da2e715c536c7641f7fb91ec4444580
-size 390108
+oid sha256:255f025ef963c2006d35050d2724c21976a3fc81ef041a980261b337ba8db61a
+size 392226