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": 76052.98851352502,
-      "required_pages_per_second_upper_estimate": 0.005261522212611454
+      "required_token_positions_per_second": 75875.88202138119,
+      "required_pages_per_second_upper_estimate": 0.005249269574541265
     },
     {
       "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": 9506.623564190628,
-      "required_pages_per_second_upper_estimate": 0.003980224765452034
+      "required_token_positions_per_second": 9484.485252672648,
+      "required_pages_per_second_upper_estimate": 0.003970955916719971
     },
     {
       "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": 19013.247128381256,
-      "required_pages_per_second_upper_estimate": 0.003959957963796473
+      "required_token_positions_per_second": 18968.970505345296,
+      "required_pages_per_second_upper_estimate": 0.003950736310871153
     },
     {
       "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": 28519.870692571883,
-      "required_pages_per_second_upper_estimate": 0.0017097227006986086
+      "required_token_positions_per_second": 28453.455758017946,
+      "required_pages_per_second_upper_estimate": 0.0017057412267818333
     },
     {
       "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": 28519.870692571883,
-      "required_pages_per_second_upper_estimate": 0.0022172945006937927
+      "required_token_positions_per_second": 28453.455758017946,
+      "required_pages_per_second_upper_estimate": 0.002212131031660648
     },
     {
       "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": 28519.870692571883,
-      "required_pages_per_second_upper_estimate": 0.3658222789929822
+      "required_token_positions_per_second": 28453.455758017946,
+      "required_pages_per_second_upper_estimate": 0.36497037952332506
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05358012771092291,
-    "producer_p95_seconds": 0.1829630560005171,
-    "gpu_update_seconds": 0.9694293602530616,
-    "producer_batches_per_second": 18.663635991971308,
-    "gpu_batches_per_second": 1.0315346749338787,
-    "producer_headroom": 18.09307669969276,
+    "producer_batch_seconds": 0.10825819981252494,
+    "producer_p95_seconds": 0.2082491770006527,
+    "gpu_update_seconds": 0.9716921640426409,
+    "producer_batches_per_second": 9.237175583297525,
+    "gpu_batches_per_second": 1.0291325143959038,
+    "producer_headroom": 8.975691132176216,
     "sustainable": true,
-    "buffer_seconds": 60.104620335689816,
+    "buffer_seconds": 60.24491417064374,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370400.27949006425,
-    "upload_bytes_per_second": 4030683.4469311656,
+    "generation_bytes_per_second": 368103.7932970754,
+    "upload_bytes_per_second": 3881562.6979871034,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.47927750135847,
+    "estimated_full_shards_per_hour": 78.98650502380558,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 5704544,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.637408425999638,
+      "bytes": 1096275,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -66,3 +66,4 @@
 {"step": 806600, "loss": 0.5344794392585754, "lr": 3e-05, "grad_norm": 0.2699462368538555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118062591116, "tokens_per_second": 150935.447145856, "data_wait_s": 1.5345999599958304e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806700, "loss": 0.536081075668335, "lr": 3e-05, "grad_norm": 0.2757851306726722, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118077248182, "tokens_per_second": 151394.7045601352, "data_wait_s": 1.7426999875169713e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806800, "loss": 0.5422860980033875, "lr": 3e-05, "grad_norm": 0.26394925850910905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118091949073, "tokens_per_second": 151874.19530871272, "data_wait_s": 1.65940000442788e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 806900, "loss": 0.5263350605964661, "lr": 3e-05, "grad_norm": 0.2738297735929711, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118106612072, "tokens_per_second": 151285.86171342045, "data_wait_s": 1.7207999917445704e-05, "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:b0ddabb958e85ca380dbc0e1df53764281137286c9a360116750406d2a53c348
-size 144972
+oid sha256:7c0b12f87b8678d518e9ab9cd37fb43e6420f0640ca30a41fb3c8d8ddd74303e
+size 147090