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": 77343.16470091294,
-      "required_pages_per_second_upper_estimate": 0.005350779594876146
+      "required_token_positions_per_second": 77505.63592195802,
+      "required_pages_per_second_upper_estimate": 0.005362019730933216
     },
     {
       "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": 9667.895587614117,
-      "required_pages_per_second_upper_estimate": 0.004047745993916642
+      "required_token_positions_per_second": 9688.204490244752,
+      "required_pages_per_second_upper_estimate": 0.004056248907349976
     },
     {
       "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": 19335.791175228234,
-      "required_pages_per_second_upper_estimate": 0.004027135382696176
+      "required_token_positions_per_second": 19376.408980489505,
+      "required_pages_per_second_upper_estimate": 0.0040355950003685415
     },
     {
       "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": 29003.686762842353,
-      "required_pages_per_second_upper_estimate": 0.0017387267353669586
+      "required_token_positions_per_second": 29064.613470734257,
+      "required_pages_per_second_upper_estimate": 0.0017423791984753818
     },
     {
       "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": 29003.686762842353,
-      "required_pages_per_second_upper_estimate": 0.002254909072075331
+      "required_token_positions_per_second": 29064.613470734257,
+      "required_pages_per_second_upper_estimate": 0.002259645855625661
     },
     {
       "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": 29003.686762842353,
-      "required_pages_per_second_upper_estimate": 0.37202815206118894
+      "required_token_positions_per_second": 29064.613470734257,
+      "required_pages_per_second_upper_estimate": 0.3728096544520242
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1139204150468629,
-    "producer_p95_seconds": 0.24800956599938218,
-    "gpu_update_seconds": 0.9532581228749453,
-    "producer_batches_per_second": 8.778057906378192,
-    "gpu_batches_per_second": 1.0490338094199347,
-    "producer_headroom": 8.367755002321648,
+    "producer_batch_seconds": 0.0626054144607906,
+    "producer_p95_seconds": 0.20970615500118583,
+    "gpu_update_seconds": 0.9512598551444466,
+    "producer_batches_per_second": 15.973059337005013,
+    "gpu_batches_per_second": 1.0512374663894046,
+    "producer_headroom": 15.194530111133039,
     "sustainable": true,
-    "buffer_seconds": 60.055261741121555,
+    "buffer_seconds": 60.88063072924458,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386799.79464909236,
-    "upload_bytes_per_second": 1911951.5839414538,
+    "generation_bytes_per_second": 387000.6660705272,
+    "upload_bytes_per_second": 1876837.605772675,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.99823169331148,
+    "estimated_full_shards_per_hour": 83.04133402430404,
     "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": 1.4949951670005248,
-      "bytes": 10484718,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.5762320020003244,
       "bytes": 8331124,
@@ -562,6 +556,12 @@
       "bytes": 11540303,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 0.9012360309970973,
+      "bytes": 1247702,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -180,3 +180,4 @@
 {"step": 663000, "loss": 0.5409803986549377, "lr": 0.0001, "grad_norm": 0.211357275389325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97045935341, "tokens_per_second": 153274.83816364087, "data_wait_s": 8.05400122771971e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 663100, "loss": 0.6096097826957703, "lr": 0.0001, "grad_norm": 0.19671272316617813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97060657989, "tokens_per_second": 153494.29424653726, "data_wait_s": 1.383099879603833e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 663200, "loss": 0.5709748864173889, "lr": 0.0001, "grad_norm": 0.1851116323886456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97075220299, "tokens_per_second": 152344.116369948, "data_wait_s": 8.39799758978188e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 663300, "loss": 0.49517443776130676, "lr": 0.0001, "grad_norm": 0.1812656144832781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97089727640, "tokens_per_second": 150520.19268278463, "data_wait_s": 1.4656001440016553e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a8926289eba84b5e6cf00f910e0eff3fd922e196ec755366e74d41b130059450
-size 387990
+oid sha256:5f6fc59071e987774bb7e4a8f40efc36d167544c08f8a1cf38072736d232e04e
+size 390108