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": 75851.77794564892,
-      "required_pages_per_second_upper_estimate": 0.006574639867755231
+      "required_token_positions_per_second": 75410.26168744279,
+      "required_pages_per_second_upper_estimate": 0.006536370357506649
     },
     {
       "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": 9481.472243206115,
-      "required_pages_per_second_upper_estimate": 0.0037823010384578407
+      "required_token_positions_per_second": 9426.282710930349,
+      "required_pages_per_second_upper_estimate": 0.0037602851088759967
     },
     {
       "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": 18962.94448641223,
-      "required_pages_per_second_upper_estimate": 0.004386743470507023
+      "required_token_positions_per_second": 18852.565421860698,
+      "required_pages_per_second_upper_estimate": 0.004361209216528213
     },
     {
       "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": 28444.416729618348,
-      "required_pages_per_second_upper_estimate": 0.0017321157425127676
+      "required_token_positions_per_second": 28278.848132791045,
+      "required_pages_per_second_upper_estimate": 0.001722033483637278
     },
     {
       "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": 28444.416729618348,
-      "required_pages_per_second_upper_estimate": 0.00218972993903106
+      "required_token_positions_per_second": 28278.848132791045,
+      "required_pages_per_second_upper_estimate": 0.002176984010124084
     },
     {
       "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": 28444.416729618348,
-      "required_pages_per_second_upper_estimate": 0.35318445844293117
+      "required_token_positions_per_second": 28278.848132791045,
+      "required_pages_per_second_upper_estimate": 0.351128650592784
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.09818565339051588,
+    "producer_batch_seconds": 0.12349865068750887,
     "producer_p95_seconds": 0.2702428470001905,
-    "gpu_update_seconds": 0.9720009470684958,
-    "producer_batches_per_second": 10.184787343856426,
-    "gpu_batches_per_second": 1.028805581945108,
-    "producer_headroom": 9.899622943919677,
+    "gpu_update_seconds": 0.9776918730979166,
+    "producer_batches_per_second": 8.097254459324581,
+    "gpu_batches_per_second": 1.0228171344325465,
+    "producer_headroom": 7.916619879287508,
     "sustainable": true,
-    "buffer_seconds": 60.264058718246744,
+    "buffer_seconds": 60.61689613207083,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296331.43734603684,
-    "upload_bytes_per_second": 4002360.23862339,
+    "generation_bytes_per_second": 304438.3868438184,
+    "upload_bytes_per_second": 4077693.2679956155,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.585828211649215,
+    "estimated_full_shards_per_hour": 65.3253908537475,
     "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": 0.884776743000657,
-      "bytes": 1180635,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.2576917689993934,
       "bytes": 6183813,
@@ -562,6 +556,12 @@
       "bytes": 2439937,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2518158220009354,
+      "bytes": 10474054,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -133,3 +133,4 @@
 {"step": 328300, "loss": 0.8265160508453846, "lr": 0.0003, "grad_norm": 0.46040005016745067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48054230333, "tokens_per_second": 150881.1920167387, "data_wait_s": 9.622001016396098e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 328400, "loss": 0.5648547895252705, "lr": 0.0003, "grad_norm": 0.1748935501663562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48068838170, "tokens_per_second": 149083.58449701086, "data_wait_s": 1.2005999451503158e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 328500, "loss": 0.5743609592318535, "lr": 0.0003, "grad_norm": 0.150383955720662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48083441066, "tokens_per_second": 148564.7945357594, "data_wait_s": 1.2335000064922497e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 328600, "loss": 0.5836680959910154, "lr": 0.0003, "grad_norm": 0.17039313687931704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48098041078, "tokens_per_second": 148264.05142376272, "data_wait_s": 1.5903999155852944e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3936c2273ebf44e028f1d958f215d91124bfa081cffe8f3c979cdf05135857aa
-size 287661
+oid sha256:c8cb5d71b8200deebc1b1c6b7cead026a9ba7a92af9fc80f892ddda80409ea51
+size 289779