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": 76656.95957691096,
-      "required_pages_per_second_upper_estimate": 0.005303306332182437
+      "required_token_positions_per_second": 76616.96723784863,
+      "required_pages_per_second_upper_estimate": 0.005300539569371093
     },
     {
       "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": 9582.11994711387,
-      "required_pages_per_second_upper_estimate": 0.004011833524438246
+      "required_token_positions_per_second": 9577.120904731079,
+      "required_pages_per_second_upper_estimate": 0.004009740529784447
     },
     {
       "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": 19164.23989422774,
-      "required_pages_per_second_upper_estimate": 0.003991405774975286
+      "required_token_positions_per_second": 19154.241809462157,
+      "required_pages_per_second_upper_estimate": 0.003989323437585841
     },
     {
       "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": 28746.359841341608,
-      "required_pages_per_second_upper_estimate": 0.0017233003793384475
+      "required_token_positions_per_second": 28731.362714193234,
+      "required_pages_per_second_upper_estimate": 0.0017224013244651897
     },
     {
       "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": 28746.359841341608,
-      "required_pages_per_second_upper_estimate": 0.002234903035790157
+      "required_token_positions_per_second": 28731.362714193234,
+      "required_pages_per_second_upper_estimate": 0.002233737075119762
     },
     {
       "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": 28746.359841341608,
-      "required_pages_per_second_upper_estimate": 0.3687274386082991
+      "required_token_positions_per_second": 28731.362714193234,
+      "required_pages_per_second_upper_estimate": 0.3685350715639003
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.3354736274847028,
-    "producer_p95_seconds": 0.34067930800665636,
-    "gpu_update_seconds": 0.9617913416723463,
-    "producer_batches_per_second": 2.9808602467435352,
-    "gpu_batches_per_second": 1.0397265567614875,
-    "producer_headroom": 2.866965576053226,
+    "producer_batch_seconds": 0.1471145790470132,
+    "producer_p95_seconds": 0.3549468139972305,
+    "gpu_update_seconds": 0.9622933751882902,
+    "producer_batches_per_second": 6.7974228419634155,
+    "gpu_batches_per_second": 1.0391841259473826,
+    "producer_headroom": 6.541114969174955,
     "sustainable": true,
-    "buffer_seconds": 60.592854525357815,
+    "buffer_seconds": 60.62448263686228,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379582.37650118564,
-    "upload_bytes_per_second": 5270481.905216138,
+    "generation_bytes_per_second": 380227.3225377675,
+    "upload_bytes_per_second": 5074672.402035379,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.4495417716663,
+    "estimated_full_shards_per_hour": 81.58793217754145,
     "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": 1.149581274999946,
-      "bytes": 13144315,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.5471423709968803,
       "bytes": 10740667,
@@ -562,6 +556,12 @@
       "bytes": 6255328,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 2.8082169360059197,
+      "bytes": 5331651,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -437,3 +437,4 @@
 {"step": 600800, "loss": 0.587449312210083, "lr": 0.0003, "grad_norm": 0.14600904392049197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87939498864, "tokens_per_second": 153686.30472890576, "data_wait_s": 9.312003385275602e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 600900, "loss": 0.5730405449867249, "lr": 0.0003, "grad_norm": 0.16492114106046663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87954036705, "tokens_per_second": 150906.22962700963, "data_wait_s": 7.586997526232153e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 601000, "loss": 0.6038550734519958, "lr": 0.0003, "grad_norm": 0.14632634401520173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87968610536, "tokens_per_second": 150951.94216182473, "data_wait_s": 1.4846998965367675e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 601100, "loss": 0.5417867302894592, "lr": 0.0003, "grad_norm": 0.15864896351462274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87983109662, "tokens_per_second": 149673.07412546175, "data_wait_s": 1.4131997886579484e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:175d694359d5e94a4bfad271fb18ef7153484344af2f0b701a2bc6082971e6f3
-size 937014
+oid sha256:271fd7874ca63ba8660c30dd64edf390c31681ee915b573e94a54c77affc345e
+size 939132