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": 75328.70594942327,
-      "required_pages_per_second_upper_estimate": 0.006529301312836245
+      "required_token_positions_per_second": 75992.34266389532,
+      "required_pages_per_second_upper_estimate": 0.006586823661274808
     },
     {
       "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": 9416.088243677908,
-      "required_pages_per_second_upper_estimate": 0.00375621838346813
+      "required_token_positions_per_second": 9499.042832986916,
+      "required_pages_per_second_upper_estimate": 0.0037893102094251297
     },
     {
       "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": 18832.176487355817,
-      "required_pages_per_second_upper_estimate": 0.004356492595363501
+      "required_token_positions_per_second": 18998.08566597383,
+      "required_pages_per_second_upper_estimate": 0.004394872763934962
     },
     {
       "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": 28248.264731033727,
-      "required_pages_per_second_upper_estimate": 0.0017201711149289654
+      "required_token_positions_per_second": 28497.12849896075,
+      "required_pages_per_second_upper_estimate": 0.0017353256127084396
     },
     {
       "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": 28248.264731033727,
-      "required_pages_per_second_upper_estimate": 0.0021746296151965313
+      "required_token_positions_per_second": 28497.12849896075,
+      "required_pages_per_second_upper_estimate": 0.0021937878369505544
     },
     {
       "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": 28248.264731033727,
-      "required_pages_per_second_upper_estimate": 0.3507489070990194
+      "required_token_positions_per_second": 28497.12849896075,
+      "required_pages_per_second_upper_estimate": 0.35383896220322025
     }
   ],
   "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": 0.12114841570308954,
-    "producer_p95_seconds": 0.38118777699855855,
-    "gpu_update_seconds": 0.9787503856697338,
-    "producer_batches_per_second": 8.254338236257247,
-    "gpu_batches_per_second": 1.0217109639407453,
-    "producer_headroom": 8.078936732185213,
+    "producer_batch_seconds": 0.11961251623421276,
+    "producer_p95_seconds": 0.3927133600009256,
+    "gpu_update_seconds": 0.9702030154023513,
+    "producer_batches_per_second": 8.360329098352084,
+    "gpu_batches_per_second": 1.0307121129543095,
+    "producer_headroom": 8.111216500977212,
     "sustainable": true,
-    "buffer_seconds": 60.6825239115235,
+    "buffer_seconds": 60.15258695494578,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366204.13577783527,
-    "upload_bytes_per_second": 6048197.354225548,
+    "generation_bytes_per_second": 366228.6466324,
+    "upload_bytes_per_second": 6128141.612331941,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.57888274194043,
+    "estimated_full_shards_per_hour": 78.58414220074654,
     "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": 1.0792534860011074,
-      "bytes": 2442489,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.0146635470009642,
       "bytes": 11041931,
@@ -562,6 +556,12 @@
       "bytes": 6084778,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.147810253998614,
+      "bytes": 8392070,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -365,3 +365,4 @@
 {"step": 76500, "loss": 0.5772983878850937, "lr": 0.0003, "grad_norm": 0.13265661268329665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11195825026, "tokens_per_second": 151144.5666849034, "data_wait_s": 8.283997885882854e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 76600, "loss": 0.5989976823329926, "lr": 0.0003, "grad_norm": 0.13564413076854506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11210538096, "tokens_per_second": 146371.2908263164, "data_wait_s": 1.207000605063513e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 76700, "loss": 0.604017112404108, "lr": 0.0003, "grad_norm": 0.1460994252826713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11225016326, "tokens_per_second": 148055.4112062978, "data_wait_s": 1.1598000128287822e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 76800, "loss": 0.6558533720672131, "lr": 0.0003, "grad_norm": 0.14504643842668982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11239515333, "tokens_per_second": 148209.24101931142, "data_wait_s": 8.47400224301964e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c4ea82f5df79976e09bc18fbda2ae93e8ba27cb7e161a02d077cb630e7322ec4
-size 782952
+oid sha256:8c7407b24aa69a8d5322013fba9db2ba04842b79ed96fc0c8b6fecf76081e1c8
+size 785070