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": 75494.70547737472,
-      "required_pages_per_second_upper_estimate": 0.005222898896775462
+      "required_token_positions_per_second": 76135.60841371072,
+      "required_pages_per_second_upper_estimate": 0.005267238049010881
     },
     {
       "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": 9436.83818467184,
-      "required_pages_per_second_upper_estimate": 0.003951007084331956
+      "required_token_positions_per_second": 9516.95105171384,
+      "required_pages_per_second_upper_estimate": 0.00398454866881554
     },
     {
       "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": 18873.67636934368,
-      "required_pages_per_second_upper_estimate": 0.003930889055417381
+      "required_token_positions_per_second": 19033.90210342768,
+      "required_pages_per_second_upper_estimate": 0.003964259850390322
     },
     {
       "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": 28310.514554015524,
-      "required_pages_per_second_upper_estimate": 0.001697172119860469
+      "required_token_positions_per_second": 28550.85315514152,
+      "required_pages_per_second_upper_estimate": 0.0017115800520221772
     },
     {
       "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": 28310.514554015524,
-      "required_pages_per_second_upper_estimate": 0.0022010179817813697
+      "required_token_positions_per_second": 28550.85315514152,
+      "required_pages_per_second_upper_estimate": 0.0022197032508811314
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28310.514554015524,
-      "required_pages_per_second_upper_estimate": 0.3631368832366892
+      "required_token_positions_per_second": 28550.85315514152,
+      "required_pages_per_second_upper_estimate": 0.366219688756449
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2488901137346602,
-    "producer_p95_seconds": 0.32697732700034976,
-    "gpu_update_seconds": 0.9765982863804377,
-    "producer_batches_per_second": 4.017837370053566,
-    "gpu_batches_per_second": 1.0239624766353994,
-    "producer_headroom": 3.9238130905495967,
+    "producer_batch_seconds": 0.10657895745299584,
+    "producer_p95_seconds": 0.3273894180019852,
+    "gpu_update_seconds": 0.968377366860614,
+    "producer_batches_per_second": 9.382715161583624,
+    "gpu_batches_per_second": 1.0326552790488108,
+    "producer_headroom": 9.08600900217751,
     "sustainable": true,
-    "buffer_seconds": 60.54909375558714,
+    "buffer_seconds": 60.03939674535807,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 537074.9003715139,
-    "upload_bytes_per_second": 5261268.416930819,
+    "generation_bytes_per_second": 559157.9493684305,
+    "upload_bytes_per_second": 5166630.792934734,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.24377115592063,
+    "estimated_full_shards_per_hour": 119.98227940358817,
     "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": 1.1021007999988797,
-      "bytes": 11824886,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.0116616440027428,
       "bytes": 2398524,
@@ -562,6 +556,12 @@
       "bytes": 7424560,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.2933890700005577,
+      "bytes": 4945280,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -369,3 +369,4 @@
 {"step": 681900, "loss": 0.5461499094963074, "lr": 0.0001, "grad_norm": 0.2342435415811837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99812491521, "tokens_per_second": 154134.3020605515, "data_wait_s": 9.36200376600027e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 682000, "loss": 0.5532965660095215, "lr": 0.0001, "grad_norm": 0.19929805827075728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99827172069, "tokens_per_second": 151754.36477470107, "data_wait_s": 1.843800419010222e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 682100, "loss": 0.5400379300117493, "lr": 0.0001, "grad_norm": 0.18625527652859342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99841673341, "tokens_per_second": 151728.03817951164, "data_wait_s": 1.9319995772093534e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 682200, "loss": 0.5104553699493408, "lr": 0.0001, "grad_norm": 0.18129928727462233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99856152253, "tokens_per_second": 151424.87195643186, "data_wait_s": 1.1705997167155147e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f74119313cbeff1779f156646d966742449417582fb5053f2eb33edb9d4349c
-size 791424
+oid sha256:43ae0633a608549e82743766dd0d911dd0725f1eaff57b3ee23b3de5cf9fe4e1
+size 793542