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": 77240.70350767874,
-      "required_pages_per_second_upper_estimate": 0.005343691091785481
+      "required_token_positions_per_second": 76986.9872028573,
+      "required_pages_per_second_upper_estimate": 0.005326138409114999
     },
     {
       "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": 9655.087938459843,
-      "required_pages_per_second_upper_estimate": 0.004042383698669875
+      "required_token_positions_per_second": 9623.373400357163,
+      "required_pages_per_second_upper_estimate": 0.0040291054838411475
     },
     {
       "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": 19310.175876919686,
-      "required_pages_per_second_upper_estimate": 0.004021800391579349
+      "required_token_positions_per_second": 19246.746800714325,
+      "required_pages_per_second_upper_estimate": 0.004008589787742998
     },
     {
       "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": 28965.263815379527,
-      "required_pages_per_second_upper_estimate": 0.0017364233383350058
+      "required_token_positions_per_second": 28870.120201071488,
+      "required_pages_per_second_upper_estimate": 0.0017307196239331262
     },
     {
       "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": 28965.263815379527,
-      "required_pages_per_second_upper_estimate": 0.0022519218569147814
+      "required_token_positions_per_second": 28870.120201071488,
+      "required_pages_per_second_upper_estimate": 0.002244524859394863
     },
     {
       "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": 28965.263815379527,
-      "required_pages_per_second_upper_estimate": 0.3715353037464826
+      "required_token_positions_per_second": 28870.120201071488,
+      "required_pages_per_second_upper_estimate": 0.3703149036193929
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14635428756236024,
-    "producer_p95_seconds": 0.3012843309988966,
-    "gpu_update_seconds": 0.9545226370532794,
-    "producer_batches_per_second": 6.832734569350481,
-    "gpu_batches_per_second": 1.0476440905446878,
-    "producer_headroom": 6.521999819421525,
+    "producer_batch_seconds": 0.08663572241403017,
+    "producer_p95_seconds": 0.3114492379972944,
+    "gpu_update_seconds": 0.9576683369324479,
+    "producer_batches_per_second": 11.542582806905244,
+    "gpu_batches_per_second": 1.044202842920699,
+    "producer_headroom": 11.053966080594012,
     "sustainable": true,
-    "buffer_seconds": 60.134926134356604,
+    "buffer_seconds": 60.333105226744216,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301221.59272829944,
+    "generation_bytes_per_second": 308716.66697147425,
     "upload_bytes_per_second": 4857587.17902014,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.63514171969163,
+    "estimated_full_shards_per_hour": 66.24341017587824,
     "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": [

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

@@ -364,3 +364,4 @@
 {"step": 796400, "loss": 0.5987265110015869, "lr": 3e-05, "grad_norm": 0.26447540313313356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116570295974, "tokens_per_second": 135728.0809670561, "data_wait_s": 1.1076997907366604e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 796500, "loss": 0.4834693968296051, "lr": 3e-05, "grad_norm": 0.26357056321213773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116585015541, "tokens_per_second": 153412.46184235357, "data_wait_s": 1.1577998520806432e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 796600, "loss": 0.520895779132843, "lr": 3e-05, "grad_norm": 0.25431231634459683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116599788305, "tokens_per_second": 153603.92232805223, "data_wait_s": 1.2486001651268452e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 796700, "loss": 0.5909894108772278, "lr": 3e-05, "grad_norm": 0.28662822347652184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116614287455, "tokens_per_second": 150561.0224878996, "data_wait_s": 1.1718999303411692e-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 sha256:4c0ad569061768d28520c2f2749da4ed061ab923cd014d3b7f847131525fd9ae
-size 780834
+oid sha256:c292ec1c11a9ca6d41fd3c342c21889d48afffdcc5e4bfb51a159d772fd8190a
+size 782952