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": 76275.23189235547,
-      "required_pages_per_second_upper_estimate": 0.00527689752523463
+      "required_token_positions_per_second": 77416.60815116968,
+      "required_pages_per_second_upper_estimate": 0.00535586058317719
     },
     {
       "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": 9534.403986544434,
-      "required_pages_per_second_upper_estimate": 0.003991855848170386
+      "required_token_positions_per_second": 9677.07601889621,
+      "required_pages_per_second_upper_estimate": 0.004051589648785249
     },
     {
       "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": 19068.807973088868,
-      "required_pages_per_second_upper_estimate": 0.003971529822511061
+      "required_token_positions_per_second": 19354.15203779242,
+      "required_pages_per_second_upper_estimate": 0.0040309594661597885
     },
     {
       "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": 28603.2119596333,
-      "required_pages_per_second_upper_estimate": 0.0017147188824041938
+      "required_token_positions_per_second": 29031.22805668863,
+      "required_pages_per_second_upper_estimate": 0.001740377793879922
     },
     {
       "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": 28603.2119596333,
-      "required_pages_per_second_upper_estimate": 0.0022237739176283254
+      "required_token_positions_per_second": 29031.22805668863,
+      "required_pages_per_second_upper_estimate": 0.0022570502865305302
     },
     {
       "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": 28603.2119596333,
-      "required_pages_per_second_upper_estimate": 0.36689129128196535
+      "required_token_positions_per_second": 29031.22805668863,
+      "required_pages_per_second_upper_estimate": 0.3723814222071116
     }
   ],
   "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": 0.08202274746102489,
-    "producer_p95_seconds": 0.22649603500030935,
-    "gpu_update_seconds": 0.9666047309308704,
-    "producer_batches_per_second": 12.191739864300137,
-    "gpu_batches_per_second": 1.0345490436788665,
-    "producer_headroom": 11.784593431111,
+    "producer_batch_seconds": 0.44138996728895563,
+    "producer_p95_seconds": 0.21887074899859726,
+    "gpu_update_seconds": 0.952353787652812,
+    "producer_batches_per_second": 2.2655702986229196,
+    "gpu_batches_per_second": 1.0500299499670367,
+    "producer_headroom": 2.15762445508725,
     "sustainable": true,
-    "buffer_seconds": 60.89609804864484,
+    "buffer_seconds": 60.95064240977997,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284166.3542818683,
-    "upload_bytes_per_second": 5451365.219749806,
+    "generation_bytes_per_second": 264457.1502220789,
+    "upload_bytes_per_second": 5469160.689951809,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.97548457471883,
+    "estimated_full_shards_per_hour": 56.74634819027686,
     "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": 1.226723518999279,
-      "bytes": 5211595,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.4460434569991776,
       "bytes": 7613200,
@@ -562,6 +556,12 @@
       "bytes": 12607883,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.1709360159984499,
+      "bytes": 6311701,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -190,3 +190,4 @@
 {"step": 544000, "loss": 0.5968501567840576, "lr": 0.0003, "grad_norm": 0.1837794646857543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79624080264, "tokens_per_second": 153831.34480019816, "data_wait_s": 9.872001101030037e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 544100, "loss": 0.5709521174430847, "lr": 0.0003, "grad_norm": 0.16342055432747546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79638776422, "tokens_per_second": 153453.21711935665, "data_wait_s": 1.3431999832391739e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 544200, "loss": 0.6047398447990417, "lr": 0.0003, "grad_norm": 0.15733616336518852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79653472941, "tokens_per_second": 153841.05317850888, "data_wait_s": 9.531999239698052e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 544300, "loss": 0.5637333989143372, "lr": 0.0003, "grad_norm": 0.15272475208395467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79668280993, "tokens_per_second": 155063.12250591084, "data_wait_s": 1.5093999536475167e-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 sha256:4b443db417f03148b31fea71e555d327039b2ffbdb6dc3f31357d7ea0a1ee980
-size 409170
+oid sha256:1ae8302a011055f1743b9270d13479dda51fd766745a436b153aee8d6a8b57d9
+size 411288