Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76379.1046660074,
-      "required_pages_per_second_upper_estimate": 0.005284083684733911
+      "required_token_positions_per_second": 76906.43338191824,
+      "required_pages_per_second_upper_estimate": 0.005320565508871819
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9547.388083250926,
-      "required_pages_per_second_upper_estimate": 0.003997292018322623
+      "required_token_positions_per_second": 9613.30417273978,
+      "required_pages_per_second_upper_estimate": 0.0040248897085071325
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19094.77616650185,
-      "required_pages_per_second_upper_estimate": 0.003976938312371672
+      "required_token_positions_per_second": 19226.60834547956,
+      "required_pages_per_second_upper_estimate": 0.004004395478604375
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28642.164249752776,
-      "required_pages_per_second_upper_estimate": 0.0017170540127201532
+      "required_token_positions_per_second": 28839.91251821934,
+      "required_pages_per_second_upper_estimate": 0.0017289087194706009
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28642.164249752776,
-      "required_pages_per_second_upper_estimate": 0.002226802286852097
+      "required_token_positions_per_second": 28839.91251821934,
+      "required_pages_per_second_upper_estimate": 0.002242176344922663
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28642.164249752776,
-      "required_pages_per_second_upper_estimate": 0.3673909294359074
+      "required_token_positions_per_second": 28839.91251821934,
+      "required_pages_per_second_upper_estimate": 0.36992743189824834
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05519175929687492,
-    "producer_p95_seconds": 0.18207385200003046,
-    "gpu_update_seconds": 0.9652901840418237,
-    "producer_batches_per_second": 18.11864692736153,
-    "gpu_batches_per_second": 1.035957908338859,
-    "producer_headroom": 17.489752027101634,
+    "producer_batch_seconds": 0.49178775737497915,
+    "producer_p95_seconds": 0.21143912100069429,
+    "gpu_update_seconds": 0.958671423934925,
+    "producer_batches_per_second": 2.0333975073672246,
+    "gpu_batches_per_second": 1.043110261799021,
+    "producer_headroom": 1.9493600838134644,
     "sustainable": true,
-    "buffer_seconds": 60.81328159463489,
+    "buffer_seconds": 60.396299707900276,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391943.1751564111,
-    "upload_bytes_per_second": 4812616.01237047,
+    "generation_bytes_per_second": 354987.1882921485,
+    "upload_bytes_per_second": 4707772.8128866805,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.10188141841172,
+    "estimated_full_shards_per_hour": 76.17198692868558,
     "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": [
@@ -334,6 +334,12 @@
       "bytes": 5745307,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9886994910002613,
+      "bytes": 1117526,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -71,3 +71,4 @@
 {"step": 727100, "loss": 0.5169442296028137, "lr": 5e-05, "grad_norm": 0.22224307388557282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106427141789, "tokens_per_second": 152708.3475259535, "data_wait_s": 1.4548000763170421e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727200, "loss": 0.50984787940979, "lr": 5e-05, "grad_norm": 0.21736101496838955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106441840632, "tokens_per_second": 151554.5667339757, "data_wait_s": 1.0786000530060846e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727300, "loss": 0.5268535017967224, "lr": 5e-05, "grad_norm": 0.23538101980784507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106456330822, "tokens_per_second": 129230.42199775409, "data_wait_s": 1.3830000170855783e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 727400, "loss": 0.5175501108169556, "lr": 5e-05, "grad_norm": 0.22082917564890717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106470834723, "tokens_per_second": 144143.5542440665, "data_wait_s": 1.110399989556754e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:69931690a67537d487ef61ad7ada8519cd29d6d1212bfe60ae48fc740c392e19
-size 155562
+oid sha256:45e2b2e8d1ec1bc15a10b52fdf7aa5d993e0b694bf032719c946ac381b000979
+size 157680