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": 77346.62747852846,
-      "required_pages_per_second_upper_estimate": 0.005351019157866338
+      "required_token_positions_per_second": 77762.59786367051,
+      "required_pages_per_second_upper_estimate": 0.005379796954294735
     },
     {
       "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": 9668.328434816058,
-      "required_pages_per_second_upper_estimate": 0.004047927218001217
+      "required_token_positions_per_second": 9720.324732958814,
+      "required_pages_per_second_upper_estimate": 0.0040696969822274125
     },
     {
       "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": 19336.656869632116,
-      "required_pages_per_second_upper_estimate": 0.004027315684010608
+      "required_token_positions_per_second": 19440.649465917628,
+      "required_pages_per_second_upper_estimate": 0.0040489745993476925
     },
     {
       "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": 29004.985304448175,
-      "required_pages_per_second_upper_estimate": 0.0017388045809534694
+      "required_token_positions_per_second": 29160.974198876444,
+      "required_pages_per_second_upper_estimate": 0.0017481558769931935
     },
     {
       "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": 29004.985304448175,
-      "required_pages_per_second_upper_estimate": 0.002255010027973503
+      "required_token_positions_per_second": 29160.974198876444,
+      "required_pages_per_second_upper_estimate": 0.002267137478392666
     },
     {
       "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": 29004.985304448175,
-      "required_pages_per_second_upper_estimate": 0.3720448083586431
+      "required_token_positions_per_second": 29160.974198876444,
+      "required_pages_per_second_upper_estimate": 0.3740456664085433
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08056049420284239,
-    "producer_p95_seconds": 0.2163603779990808,
-    "gpu_update_seconds": 0.9532154458895703,
-    "producer_batches_per_second": 12.413032093399414,
-    "gpu_batches_per_second": 1.0490807763472285,
-    "producer_headroom": 11.832293921751267,
+    "producer_batch_seconds": 0.09064698599985377,
+    "producer_p95_seconds": 0.21593819400004577,
+    "gpu_update_seconds": 0.948116472770833,
+    "producer_batches_per_second": 11.031806396757784,
+    "gpu_batches_per_second": 1.0547227357811213,
+    "producer_headroom": 10.459437369184702,
     "sustainable": true,
-    "buffer_seconds": 60.05257309104293,
+    "buffer_seconds": 60.67945425733331,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351206.2791259869,
-    "upload_bytes_per_second": 5687277.982225071,
+    "generation_bytes_per_second": 351230.78988264914,
+    "upload_bytes_per_second": 5482438.459894346,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.36069183668809,
+    "estimated_full_shards_per_hour": 75.36595127448659,
     "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": 0.8962630369987892,
-      "bytes": 1124581,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.1323460120001982,
       "bytes": 7094094,
@@ -562,6 +556,12 @@
       "bytes": 12177821,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 3.9100393570006418,
+      "bytes": 2234803,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -162,3 +162,4 @@
 {"step": 541200, "loss": 0.5968453884124756, "lr": 0.0003, "grad_norm": 0.18736173766674702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79213582722, "tokens_per_second": 152406.80808195958, "data_wait_s": 1.8074999388772994e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 541300, "loss": 0.5071848630905151, "lr": 0.0003, "grad_norm": 0.1660437392540106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79228255960, "tokens_per_second": 153362.80730092418, "data_wait_s": 1.294399771722965e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 541400, "loss": 0.5476915240287781, "lr": 0.0003, "grad_norm": 0.1653532639298301, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79243004314, "tokens_per_second": 154845.0489894277, "data_wait_s": 1.3023000065004453e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 541500, "loss": 0.5443552732467651, "lr": 0.0003, "grad_norm": 0.15322251732361983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79257695084, "tokens_per_second": 138901.01671732092, "data_wait_s": 1.4998997357906774e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3655317edb37fcee04e1c4d676b3ddb89be618c0016f5a64dfe72e4e388795f5
-size 349866
+oid sha256:3af130106bf981b1f607506639a852b30d7c6e6226b506fe8b0c3fb40b38c121
+size 351984