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": 74997.99293920214,
-      "required_pages_per_second_upper_estimate": 0.0065006359472947394
+      "required_token_positions_per_second": 74906.44612878068,
+      "required_pages_per_second_upper_estimate": 0.0064927009017906765
     },
     {
       "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": 9374.749117400268,
-      "required_pages_per_second_upper_estimate": 0.0037397275879209623
+      "required_token_positions_per_second": 9363.305766097585,
+      "required_pages_per_second_upper_estimate": 0.003735162663992973
     },
     {
       "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": 18749.498234800536,
-      "required_pages_per_second_upper_estimate": 0.0043373664367223844
+      "required_token_positions_per_second": 18726.61153219517,
+      "required_pages_per_second_upper_estimate": 0.0043320720008668415
     },
     {
       "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": 28124.247352200808,
-      "required_pages_per_second_upper_estimate": 0.0017126191072269412
+      "required_token_positions_per_second": 28089.917298292752,
+      "required_pages_per_second_upper_estimate": 0.0017105285870598133
     },
     {
       "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": 28124.247352200808,
-      "required_pages_per_second_upper_estimate": 0.002165082414071896
+      "required_token_positions_per_second": 28089.917298292752,
+      "required_pages_per_second_upper_estimate": 0.0021624395914903827
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28124.247352200808,
-      "required_pages_per_second_upper_estimate": 0.3492090263133815
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28089.917298292752,
+      "required_pages_per_second_upper_estimate": 0.3487827619391429
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13015100266363788,
-    "producer_p95_seconds": 0.4057546660042135,
-    "gpu_update_seconds": 0.9830663076513037,
-    "producer_batches_per_second": 7.683382990021206,
-    "gpu_batches_per_second": 1.0172253816623555,
-    "producer_headroom": 7.553274946270981,
+    "producer_batch_seconds": 0.12145421442204452,
+    "producer_p95_seconds": 0.4090576239977963,
+    "gpu_update_seconds": 0.9842677607911786,
+    "producer_batches_per_second": 8.23355537523855,
+    "gpu_batches_per_second": 1.0159836985782968,
+    "producer_headroom": 8.10402311253622,
     "sustainable": true,
-    "buffer_seconds": 60.95011107438083,
+    "buffer_seconds": 60.040333408261894,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266424.82409696246,
-    "upload_bytes_per_second": 5642544.469684359,
+    "generation_bytes_per_second": 280850.14091640327,
+    "upload_bytes_per_second": 5659773.199429384,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.168565198723364,
+    "estimated_full_shards_per_hour": 60.26390238398622,
     "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": 2.290356701996643,
-      "bytes": 13205046,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.0317470599984517,
       "bytes": 1158559,
@@ -562,6 +556,12 @@
       "bytes": 9783282,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.0601626389980083,
+      "bytes": 7547567,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -401,3 +401,4 @@
 {"step": 235100, "loss": 0.6112591940909624, "lr": 0.0003, "grad_norm": 0.14457853247116872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34412172596, "tokens_per_second": 123930.859132861, "data_wait_s": 1.5904995962046087e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 235200, "loss": 0.5915999626740813, "lr": 0.0003, "grad_norm": 0.139044980268077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34426840209, "tokens_per_second": 148063.79049555262, "data_wait_s": 1.612499909242615e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 235300, "loss": 0.5560144484043121, "lr": 0.0003, "grad_norm": 0.12675935574134922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34441629609, "tokens_per_second": 149851.10830901898, "data_wait_s": 1.2976001016795635e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 235400, "loss": 0.6404387168586254, "lr": 0.0003, "grad_norm": 0.16956014142682727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34456249059, "tokens_per_second": 148788.3291364617, "data_wait_s": 1.3794000551570207e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:665a0722d21716c72cf0e7c865bb8f1f9102a95d445da4423fd581ad10c365ca
-size 859983
+oid sha256:59123e93d6fa8e7ea0c157fecd517788179d26b7c4632574126209705f0c5231
+size 862101