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": 77417.17594884554,
-      "required_pages_per_second_upper_estimate": 0.005355899864737878
+      "required_token_positions_per_second": 77318.16052700972,
+      "required_pages_per_second_upper_estimate": 0.005349049748107845
     },
     {
       "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": 9677.146993605693,
-      "required_pages_per_second_upper_estimate": 0.004051619364413989
+      "required_token_positions_per_second": 9664.770065876215,
+      "required_pages_per_second_upper_estimate": 0.004046437403233297
     },
     {
       "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": 19354.293987211386,
-      "required_pages_per_second_upper_estimate": 0.004030989030480301
+      "required_token_positions_per_second": 19329.54013175243,
+      "required_pages_per_second_upper_estimate": 0.004025833455191268
     },
     {
       "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": 29031.440980817082,
-      "required_pages_per_second_upper_estimate": 0.0017403905583563066
+      "required_token_positions_per_second": 28994.31019762865,
+      "required_pages_per_second_upper_estimate": 0.0017381646245995832
     },
     {
       "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": 29031.440980817082,
-      "required_pages_per_second_upper_estimate": 0.002257066840444963
+      "required_token_positions_per_second": 28994.31019762865,
+      "required_pages_per_second_upper_estimate": 0.0022541800853731176
     },
     {
       "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": 29031.440980817082,
-      "required_pages_per_second_upper_estimate": 0.3723841533692113
+      "required_token_positions_per_second": 28994.31019762865,
+      "required_pages_per_second_upper_estimate": 0.3719078795503989
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11906483204683127,
-    "producer_p95_seconds": 0.3160398109976086,
-    "gpu_update_seconds": 0.9523468028427797,
-    "producer_batches_per_second": 8.39878562636089,
-    "gpu_batches_per_second": 1.0500376512158955,
-    "producer_headroom": 7.998556639026687,
+    "producer_batch_seconds": 0.6383114484142425,
+    "producer_p95_seconds": 0.32732526100153336,
+    "gpu_update_seconds": 0.953566400150511,
+    "producer_batches_per_second": 1.5666333456564199,
+    "gpu_batches_per_second": 1.0486946686063603,
+    "producer_headroom": 1.4938889197733436,
     "sustainable": true,
-    "buffer_seconds": 60.9501953819379,
+    "buffer_seconds": 60.07468320948219,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398849.2640936334,
-    "upload_bytes_per_second": 5259152.107175771,
+    "generation_bytes_per_second": 326292.670501304,
+    "upload_bytes_per_second": 5242086.609836982,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.58376733881713,
+    "estimated_full_shards_per_hour": 70.0148113849577,
     "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": 1.1014808299987635,
-      "bytes": 5892940,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.8753045230005227,
       "bytes": 5867523,
@@ -562,6 +556,12 @@
       "bytes": 232525,
       "pages": 6,
       "shards": 1
+    },
+    {
+      "seconds": 1.743612624995876,
+      "bytes": 7907888,
+      "pages": 23,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 474900, "loss": 0.5995561480522156, "lr": 0.0003, "grad_norm": 0.15530859736052954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69510239205, "tokens_per_second": 153163.09632025065, "data_wait_s": 1.4687000657431781e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475000, "loss": 0.6206049919128418, "lr": 0.0003, "grad_norm": 0.16388562270623272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69524850745, "tokens_per_second": 151958.12900340825, "data_wait_s": 1.5634002920705825e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475100, "loss": 0.5530853271484375, "lr": 0.0003, "grad_norm": 0.15374130826137225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69539550679, "tokens_per_second": 136254.72319083425, "data_wait_s": 1.0569994628895074e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 475200, "loss": 0.6070004105567932, "lr": 0.0003, "grad_norm": 0.15388277254052188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69554224058, "tokens_per_second": 134917.34085297797, "data_wait_s": 8.978000550996512e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4eb9db34c8683fe7ff4038964f22981c7b035293ccb5e0734b97a99ea4ae4503
-size 744828
+oid sha256:86fafd44c066525c1cf71243e0d35d499c34b424754efe72a4f85356a1bbc562
+size 746946