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": 76901.51707841772,
-      "required_pages_per_second_upper_estimate": 0.005320225387588257
+      "required_token_positions_per_second": 76427.41686507025,
+      "required_pages_per_second_upper_estimate": 0.005287426034764825
     },
     {
       "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": 9612.689634802215,
-      "required_pages_per_second_upper_estimate": 0.0040246324143056855
+      "required_token_positions_per_second": 9553.427108133781,
+      "required_pages_per_second_upper_estimate": 0.003999820431931895
     },
     {
       "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": 19225.37926960443,
-      "required_pages_per_second_upper_estimate": 0.0040041394945124955
+      "required_token_positions_per_second": 19106.854216267562,
+      "required_pages_per_second_upper_estimate": 0.0039794538516182785
     },
     {
       "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": 28838.068904406646,
-      "required_pages_per_second_upper_estimate": 0.0017287981976375653
+      "required_token_positions_per_second": 28660.281324401345,
+      "required_pages_per_second_upper_estimate": 0.0017181401036821637
     },
     {
       "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": 28838.068904406646,
-      "required_pages_per_second_upper_estimate": 0.0022420330120579263
+      "required_token_positions_per_second": 28660.281324401345,
+      "required_pages_per_second_upper_estimate": 0.002228210809717434
     },
     {
       "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": 28838.068904406646,
-      "required_pages_per_second_upper_estimate": 0.3699037839997774
+      "required_token_positions_per_second": 28660.281324401345,
+      "required_pages_per_second_upper_estimate": 0.36762331581690005
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08377278296066493,
-    "producer_p95_seconds": 0.2393416380000417,
-    "gpu_update_seconds": 0.9587327116683324,
-    "producer_batches_per_second": 11.937051207544876,
-    "gpu_batches_per_second": 1.043043580165171,
-    "producer_headroom": 11.44444147353324,
+    "producer_batch_seconds": 0.11754560235192457,
+    "producer_p95_seconds": 0.2502775129978545,
+    "gpu_update_seconds": 0.9646799934395798,
+    "producer_batches_per_second": 8.507336556973517,
+    "gpu_batches_per_second": 1.0366131844763218,
+    "producer_headroom": 8.206857373969509,
     "sustainable": true,
-    "buffer_seconds": 60.40016083510494,
+    "buffer_seconds": 60.774839586693524,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316585.4341655813,
-    "upload_bytes_per_second": 3791565.3371887053,
+    "generation_bytes_per_second": 299063.31177955,
+    "upload_bytes_per_second": 3800165.1127890213,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.93186444020824,
+    "estimated_full_shards_per_hour": 64.1720248702991,
     "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": 1.6944532489997073,
-      "bytes": 1160058,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 2.4445176089993765,
       "bytes": 7341476,
@@ -562,6 +556,12 @@
       "bytes": 5939964,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2798854280008527,
+      "bytes": 545338,
+      "pages": 18,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -180,3 +180,4 @@
 {"step": 818000, "loss": 0.5703120231628418, "lr": 3e-05, "grad_norm": 0.2724048763442829, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119732351605, "tokens_per_second": 151184.33777037638, "data_wait_s": 1.6276000678772107e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 818100, "loss": 0.5537112355232239, "lr": 3e-05, "grad_norm": 0.28214828228673816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119746897283, "tokens_per_second": 151022.87739474027, "data_wait_s": 1.4263001503422856e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 818200, "loss": 0.5110453367233276, "lr": 3e-05, "grad_norm": 0.281281517734612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119761321195, "tokens_per_second": 126540.26405657713, "data_wait_s": 1.0697000107029453e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 818300, "loss": 0.5910853147506714, "lr": 3e-05, "grad_norm": 0.2962499751839353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119775860827, "tokens_per_second": 139799.13285007075, "data_wait_s": 8.857998182065785e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb3aa759a6f957aead43732793387b6fb132469fe182c6f338aacc9050f1c63e
-size 387990
+oid sha256:ca1f90420ffa6ab41ccd7da9a60b041b0d9b438dda5ca5b5791ff37527c57a60
+size 390108