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": 77348.08219018958,
-      "required_pages_per_second_upper_estimate": 0.005351119798194451
+      "required_token_positions_per_second": 77516.08515030619,
+      "required_pages_per_second_upper_estimate": 0.005362742632795891
     },
     {
       "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": 9668.510273773698,
-      "required_pages_per_second_upper_estimate": 0.004048003350175551
+      "required_token_positions_per_second": 9689.510643788273,
+      "required_pages_per_second_upper_estimate": 0.004056795766563061
     },
     {
       "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": 19337.020547547396,
-      "required_pages_per_second_upper_estimate": 0.004027391428529529
+      "required_token_positions_per_second": 19379.021287576546,
+      "required_pages_per_second_upper_estimate": 0.004036139075043602
     },
     {
       "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": 29005.530821321096,
-      "required_pages_per_second_upper_estimate": 0.0017388372838570449
+      "required_token_positions_per_second": 29068.53193136482,
+      "required_pages_per_second_upper_estimate": 0.0017426141042070412
     },
     {
       "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": 29005.530821321096,
-      "required_pages_per_second_upper_estimate": 0.0022550524395108973
+      "required_token_positions_per_second": 29068.53193136482,
+      "required_pages_per_second_upper_estimate": 0.0022599504987042
     },
     {
       "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": 29005.530821321096,
-      "required_pages_per_second_upper_estimate": 0.37205180566335855
+      "required_token_positions_per_second": 29068.53193136482,
+      "required_pages_per_second_upper_estimate": 0.37285991625767784
     }
   ],
   "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": 0.07833567796878071,
-    "producer_p95_seconds": 0.2085137370013399,
-    "gpu_update_seconds": 0.9531975184428201,
-    "producer_batches_per_second": 12.765575353781097,
-    "gpu_batches_per_second": 1.049100507136903,
-    "producer_headroom": 12.168114748718967,
+    "producer_batch_seconds": 0.05790433517186955,
+    "producer_p95_seconds": 0.19817051400059427,
+    "gpu_update_seconds": 0.9511316245788087,
+    "producer_batches_per_second": 17.26986411348712,
+    "gpu_batches_per_second": 1.0513791931193874,
+    "producer_headroom": 16.425913910516275,
     "sustainable": true,
-    "buffer_seconds": 60.051443661897665,
+    "buffer_seconds": 60.87242397304376,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276787.44919160224,
-    "upload_bytes_per_second": 3393543.606563083,
+    "generation_bytes_per_second": 276382.74996421445,
+    "upload_bytes_per_second": 3351577.3811387233,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.392143314466956,
+    "estimated_full_shards_per_hour": 59.30530428118539,
     "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": 0.9827144040000348,
-      "bytes": 5877076,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.1320399030000772,
       "bytes": 865700,
@@ -562,6 +556,12 @@
       "bytes": 10017417,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.1130377059998864,
+      "bytes": 986874,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -128,3 +128,4 @@
 {"step": 697800, "loss": 0.5559768676757812, "lr": 8e-05, "grad_norm": 0.19608597554671559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102139096706, "tokens_per_second": 152363.87423358994, "data_wait_s": 1.2340000466792844e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 697900, "loss": 0.581475019454956, "lr": 8e-05, "grad_norm": 0.20897377201278122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102153670618, "tokens_per_second": 141078.9718163793, "data_wait_s": 1.1224999980186112e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 698000, "loss": 0.534964382648468, "lr": 8e-05, "grad_norm": 0.19932546934267295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102168138799, "tokens_per_second": 151464.36256755728, "data_wait_s": 1.2217000403325073e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 698100, "loss": 0.5334374904632568, "lr": 8e-05, "grad_norm": 0.2169531830134169, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102182812897, "tokens_per_second": 149720.55519985966, "data_wait_s": 2.351700095459819e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d43caca7a73e920fc92e2e7835f34f59e19ae5b1a6470542fc54d99651167980
-size 277071
+oid sha256:ec70b02ee89761f07faad6891222429337b6646840a1157f431a59023ccc98e6
+size 279189