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": 75731.21103212437,
-      "required_pages_per_second_upper_estimate": 0.0065641894332649706
+      "required_token_positions_per_second": 75969.77179854602,
+      "required_pages_per_second_upper_estimate": 0.006584867275871665
     },
     {
       "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": 9466.401379015546,
-      "required_pages_per_second_upper_estimate": 0.0037762890454027228
+      "required_token_positions_per_second": 9496.221474818252,
+      "required_pages_per_second_upper_estimate": 0.0037881847274685862
     },
     {
       "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": 18932.80275803109,
-      "required_pages_per_second_upper_estimate": 0.004379770712122348
+      "required_token_positions_per_second": 18992.442949636505,
+      "required_pages_per_second_upper_estimate": 0.004393567420818738
     },
     {
       "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": 28399.20413704664,
-      "required_pages_per_second_upper_estimate": 0.0017293625328372911
+      "required_token_positions_per_second": 28488.664424454757,
+      "required_pages_per_second_upper_estimate": 0.0017348101949785917
     },
     {
       "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": 28399.20413704664,
-      "required_pages_per_second_upper_estimate": 0.002186249348498423
+      "required_token_positions_per_second": 28488.664424454757,
+      "required_pages_per_second_upper_estimate": 0.0021931362490650256
     },
     {
       "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": 28399.20413704664,
-      "required_pages_per_second_upper_estimate": 0.352623069360004
+      "required_token_positions_per_second": 28488.664424454757,
+      "required_pages_per_second_upper_estimate": 0.35373386672529095
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.19920030615637074,
-    "producer_p95_seconds": 0.24541962999865063,
-    "gpu_update_seconds": 0.9735484088419684,
-    "producer_batches_per_second": 5.0200726057870995,
-    "gpu_batches_per_second": 1.027170288521652,
-    "producer_headroom": 4.887283697635184,
+    "producer_batch_seconds": 0.18019159153911346,
+    "producer_p95_seconds": 0.2058232279996446,
+    "gpu_update_seconds": 0.9704912658617605,
+    "producer_batches_per_second": 5.54964852387651,
+    "gpu_batches_per_second": 1.0304059760002444,
+    "producer_headroom": 5.385885421024764,
     "sustainable": true,
-    "buffer_seconds": 60.36000134820204,
+    "buffer_seconds": 60.17045848342915,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 442214.1127743015,
-    "upload_bytes_per_second": 5984337.118994301,
+    "generation_bytes_per_second": 441637.210246808,
+    "upload_bytes_per_second": 5841700.966734531,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.88885438367637,
+    "estimated_full_shards_per_hour": 94.76506453087978,
     "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": [
@@ -430,6 +430,18 @@
       "bytes": 7549095,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0992059410018555,
+      "bytes": 4757047,
+      "pages": 31,
+      "shards": 2
+    },
+    {
+      "seconds": 0.9332335270009935,
+      "bytes": 581892,
+      "pages": 17,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -98,3 +98,4 @@
 {"step": 94800, "loss": 0.6088749971240759, "lr": 0.0003, "grad_norm": 0.13870514125347966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13873367436, "tokens_per_second": 149842.19417131858, "data_wait_s": 9.170002158498392e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 94900, "loss": 0.6596100553870201, "lr": 0.0003, "grad_norm": 0.14721265667715122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13888128543, "tokens_per_second": 151242.02691629334, "data_wait_s": 1.3546003174269572e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 95000, "loss": 0.6396308094263077, "lr": 0.0003, "grad_norm": 0.143320235435905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13902633353, "tokens_per_second": 148697.1748974744, "data_wait_s": 1.7082998965634033e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 95100, "loss": 0.6108560711145401, "lr": 0.0003, "grad_norm": 0.13584040699080668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13917151076, "tokens_per_second": 131626.97595572475, "data_wait_s": 1.3256998499855399e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cacd16700d0a0856ccf98ecbb551875623f3d2c846fad699f11de75c93495d74
-size 213531
+oid sha256:c5cf460de7040f98cd0b5ebc6e7f981d69b87b6aa3c9ec19dd31af49afbfb018
+size 215649