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": 76823.22785327848,
-      "required_pages_per_second_upper_estimate": 0.005314809157337095
+      "required_token_positions_per_second": 76818.16959106401,
+      "required_pages_per_second_upper_estimate": 0.005314459215020313
     },
     {
       "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": 9602.90348165981,
-      "required_pages_per_second_upper_estimate": 0.004020535156343077
+      "required_token_positions_per_second": 9602.271198883001,
+      "required_pages_per_second_upper_estimate": 0.004020270432747992
     },
     {
       "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": 19205.80696331962,
-      "required_pages_per_second_upper_estimate": 0.004000063099269797
+      "required_token_positions_per_second": 19204.542397766003,
+      "required_pages_per_second_upper_estimate": 0.003999799723613815
     },
     {
       "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": 28808.710444979435,
-      "required_pages_per_second_upper_estimate": 0.001727038202822676
+      "required_token_positions_per_second": 28806.813596649008,
+      "required_pages_per_second_upper_estimate": 0.0017269244896615866
     },
     {
       "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": 28808.710444979435,
-      "required_pages_per_second_upper_estimate": 0.0022397505209716764
+      "required_token_positions_per_second": 28806.813596649008,
+      "required_pages_per_second_upper_estimate": 0.00223960304935734
     },
     {
       "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": 28808.710444979435,
-      "required_pages_per_second_upper_estimate": 0.3695272052049029
+      "required_token_positions_per_second": 28806.813596649008,
+      "required_pages_per_second_upper_estimate": 0.36950287447119723
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.4716353147346126,
-    "producer_p95_seconds": 0.33666720500332303,
-    "gpu_update_seconds": 0.9597097396221109,
-    "producer_batches_per_second": 2.1202822790373452,
-    "gpu_batches_per_second": 1.0419817145898234,
-    "producer_headroom": 2.0348555539403064,
+    "producer_batch_seconds": 0.15495285810953874,
+    "producer_p95_seconds": 0.34062968799844384,
+    "gpu_update_seconds": 0.9597729338317443,
+    "producer_batches_per_second": 6.453575701669752,
+    "gpu_batches_per_second": 1.0419131075176868,
+    "producer_headroom": 6.193967284896836,
     "sustainable": true,
-    "buffer_seconds": 60.461713596192986,
+    "buffer_seconds": 60.465694831399894,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323622.9729696149,
-    "upload_bytes_per_second": 4587267.528158256,
+    "generation_bytes_per_second": 322548.29542508785,
+    "upload_bytes_per_second": 4640412.647905926,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.44195644203506,
+    "estimated_full_shards_per_hour": 69.21135565819242,
     "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": 1.7779917830011982,
-      "bytes": 7496140,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.6659519919994636,
       "bytes": 6121395,
@@ -562,6 +556,12 @@
       "bytes": 11904346,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.4954820010025287,
+      "bytes": 11186044,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -401,3 +401,4 @@
 {"step": 800100, "loss": 0.5328884124755859, "lr": 3e-05, "grad_norm": 0.26596277544685065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117111234819, "tokens_per_second": 144089.23081662506, "data_wait_s": 1.2935997801832855e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 800200, "loss": 0.5206485986709595, "lr": 3e-05, "grad_norm": 0.25603326245819596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117125930630, "tokens_per_second": 152573.85255618553, "data_wait_s": 9.213996236212552e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 800300, "loss": 0.5441808700561523, "lr": 3e-05, "grad_norm": 0.27475873093016856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117140665293, "tokens_per_second": 153062.38374156097, "data_wait_s": 1.7004000255838037e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 800400, "loss": 0.5579537153244019, "lr": 3e-05, "grad_norm": 0.2773674184649962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117155369920, "tokens_per_second": 153523.41712702133, "data_wait_s": 1.5716002963017672e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2e6d4af175bdee60727a3b70853c1c24741a43fe832906816c0b77531f40e32
-size 859983
+oid sha256:1deea560265b4d2ab6b5315ec75c6eab11f3be1ac44807a98ca5c0e6ba9e8e92
+size 862101