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": 77131.9158538709,
-      "required_pages_per_second_upper_estimate": 0.005336164909472921
+      "required_token_positions_per_second": 77005.2866150109,
+      "required_pages_per_second_upper_estimate": 0.005327404404908272
     },
     {
       "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": 9641.489481733863,
-      "required_pages_per_second_upper_estimate": 0.004036690308806774
+      "required_token_positions_per_second": 9625.660826876363,
+      "required_pages_per_second_upper_estimate": 0.0040300631815578395
     },
     {
       "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": 19282.978963467725,
-      "required_pages_per_second_upper_estimate": 0.004016135991738144
+      "required_token_positions_per_second": 19251.321653752726,
+      "required_pages_per_second_upper_estimate": 0.004009542608983909
     },
     {
       "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": 28924.468445201586,
-      "required_pages_per_second_upper_estimate": 0.0017339777181837619
+      "required_token_positions_per_second": 28876.98248062909,
+      "required_pages_per_second_upper_estimate": 0.0017311310071145103
     },
     {
       "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": 28924.468445201586,
-      "required_pages_per_second_upper_estimate": 0.002248750196323316
+      "required_token_positions_per_second": 28876.98248062909,
+      "required_pages_per_second_upper_estimate": 0.002245058371446489
     },
     {
       "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": 28924.468445201586,
-      "required_pages_per_second_upper_estimate": 0.3710120245404957
+      "required_token_positions_per_second": 28876.98248062909,
+      "required_pages_per_second_upper_estimate": 0.37040292557341603
     }
   ],
   "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": 0.09374171767188955,
-    "producer_p95_seconds": 0.22251231400150573,
-    "gpu_update_seconds": 0.9558689056768701,
-    "producer_batches_per_second": 10.667609094812558,
-    "gpu_batches_per_second": 1.0461685635561917,
-    "producer_headroom": 10.196835831647107,
+    "producer_batch_seconds": 0.08088870408573712,
+    "producer_p95_seconds": 0.22170357599679846,
+    "gpu_update_seconds": 0.9574407581729324,
+    "producer_batches_per_second": 12.362665607055105,
+    "gpu_batches_per_second": 1.0444510445829387,
+    "producer_headroom": 11.836519931857275,
     "sustainable": true,
-    "buffer_seconds": 60.219741057642814,
+    "buffer_seconds": 60.318767764894744,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379858.94826576445,
-    "upload_bytes_per_second": 5043522.72989217,
+    "generation_bytes_per_second": 409352.7608143982,
+    "upload_bytes_per_second": 5101067.746899259,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.50888763408375,
+    "estimated_full_shards_per_hour": 87.83757322620353,
     "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": 0.855230520999612,
-      "bytes": 838138,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.1489447340009065,
       "bytes": 10403004,
@@ -562,6 +556,12 @@
       "bytes": 2415324,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.996803005997208,
+      "bytes": 6353091,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -172,3 +172,4 @@
 {"step": 777200, "loss": 0.5374730825424194, "lr": 3e-05, "grad_norm": 0.24542080249639392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113759763064, "tokens_per_second": 150248.51303089628, "data_wait_s": 9.609000699128956e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 777300, "loss": 0.5130161643028259, "lr": 3e-05, "grad_norm": 0.2538126064334016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113774514532, "tokens_per_second": 153109.124234188, "data_wait_s": 1.2678003258770332e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 777400, "loss": 0.5885140299797058, "lr": 3e-05, "grad_norm": 0.2971185917362797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113789274191, "tokens_per_second": 153123.52556319867, "data_wait_s": 1.220199919771403e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 777500, "loss": 0.5324936509132385, "lr": 3e-05, "grad_norm": 0.24588803437544945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113803985646, "tokens_per_second": 153262.4781311956, "data_wait_s": 1.1507996532600373e-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 sha256:06416552d75ff561a212908ea76dfad1b4ef01fc3c05e50759f97aff3da30494
-size 371046
+oid sha256:ba989cf49973f7ec37c9a483f784801d7ab253037559c803ae9aca9cf8d80c96
+size 373164