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": 76745.67127816028,
-      "required_pages_per_second_upper_estimate": 0.005309443613514371
+      "required_token_positions_per_second": 77050.07757932403,
+      "required_pages_per_second_upper_estimate": 0.005330503147748809
     },
     {
       "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": 9593.208909770035,
-      "required_pages_per_second_upper_estimate": 0.00401647624153851
+      "required_token_positions_per_second": 9631.259697415504,
+      "required_pages_per_second_upper_estimate": 0.004032407311734865
     },
     {
       "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": 19186.41781954007,
-      "required_pages_per_second_upper_estimate": 0.003996024851946623
+      "required_token_positions_per_second": 19262.519394831008,
+      "required_pages_per_second_upper_estimate": 0.004011874803146224
     },
     {
       "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": 28779.626729310105,
-      "required_pages_per_second_upper_estimate": 0.001725294678476562
+      "required_token_positions_per_second": 28893.779092246514,
+      "required_pages_per_second_upper_estimate": 0.0017321379383340339
     },
     {
       "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": 28779.626729310105,
-      "required_pages_per_second_upper_estimate": 0.002237489390004132
+      "required_token_positions_per_second": 28893.779092246514,
+      "required_pages_per_second_upper_estimate": 0.002246364234119257
     },
     {
       "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": 28779.626729310105,
-      "required_pages_per_second_upper_estimate": 0.36915415052795764
+      "required_token_positions_per_second": 28893.779092246514,
+      "required_pages_per_second_upper_estimate": 0.37061837447244794
     }
   ],
   "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": 0.47901179265627647,
-    "producer_p95_seconds": 0.14705047899997226,
-    "gpu_update_seconds": 0.9606795897683544,
-    "producer_batches_per_second": 2.087631276162689,
-    "gpu_batches_per_second": 1.0409297862163667,
-    "producer_headroom": 2.005544757971559,
+    "producer_batch_seconds": 0.08908060844532528,
+    "producer_p95_seconds": 0.13864404100013417,
+    "gpu_update_seconds": 0.9568841760619395,
+    "producer_batches_per_second": 11.22578771578291,
+    "gpu_batches_per_second": 1.045058560917481,
+    "producer_headroom": 10.741778629063173,
     "sustainable": true,
-    "buffer_seconds": 60.52281415540633,
+    "buffer_seconds": 60.28370309190219,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 264215.6544306468,
-    "upload_bytes_per_second": 4702605.281889259,
+    "generation_bytes_per_second": 265296.9491017718,
+    "upload_bytes_per_second": 4806223.740536115,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.69452881516984,
+    "estimated_full_shards_per_hour": 56.926549480341585,
     "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": [
@@ -280,6 +280,12 @@
       "bytes": 6082079,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0907009550001021,
+      "bytes": 7267703,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -17,3 +17,4 @@
 {"step": 606100, "loss": 0.5854063630104065, "lr": 0.0001, "grad_norm": 0.13425116211854862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88715194408, "tokens_per_second": 153708.0282001785, "data_wait_s": 8.715000149095431e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 606200, "loss": 0.5686312913894653, "lr": 0.0001, "grad_norm": 0.14727125806159375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88729746436, "tokens_per_second": 150734.31198214568, "data_wait_s": 1.3697999747819267e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 606300, "loss": 0.5561347603797913, "lr": 0.0001, "grad_norm": 0.14712437311455911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88744266322, "tokens_per_second": 123861.64412074059, "data_wait_s": 1.2516999959188979e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 606400, "loss": 0.5870782136917114, "lr": 0.0001, "grad_norm": 0.1859295387640888, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88758968024, "tokens_per_second": 152193.9336136884, "data_wait_s": 2.0631000097637298e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e2f5ba4a53960c2486116098178c1db9559a417a1f030598f1b7bcc271acd041
-size 41190
+oid sha256:7028b866ce54ece4892511b2248736c92819884ddbd695790b99c36d6d5567cb
+size 43308