Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75769.57102539633,
-      "required_pages_per_second_upper_estimate": 0.006567514380259271
+      "required_token_positions_per_second": 74951.0434251579,
+      "required_pages_per_second_upper_estimate": 0.00649656648240984
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9471.196378174542,
-      "required_pages_per_second_upper_estimate": 0.0037782018422588726
+      "required_token_positions_per_second": 9368.880428144737,
+      "required_pages_per_second_upper_estimate": 0.0037373864800322072
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18942.392756349083,
-      "required_pages_per_second_upper_estimate": 0.004381989189455009
+      "required_token_positions_per_second": 18737.760856289475,
+      "required_pages_per_second_upper_estimate": 0.004334651200774657
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28413.589134523627,
-      "required_pages_per_second_upper_estimate": 0.0017302385037113908
+      "required_token_positions_per_second": 28106.641284434216,
+      "required_pages_per_second_upper_estimate": 0.0017115469900718543
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28413.589134523627,
-      "required_pages_per_second_upper_estimate": 0.0021873567454244233
+      "required_token_positions_per_second": 28106.641284434216,
+      "required_pages_per_second_upper_estimate": 0.0021637270502385094
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28413.589134523627,
-      "required_pages_per_second_upper_estimate": 0.35280168288517855
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28106.641284434216,
+      "required_pages_per_second_upper_estimate": 0.34899041787543883
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1073139328513264,
-    "producer_p95_seconds": 0.3570089419990836,
-    "gpu_update_seconds": 0.9730555287859285,
-    "producer_batches_per_second": 9.318454495423332,
-    "gpu_batches_per_second": 1.0276905792290085,
-    "producer_headroom": 9.067373666511763,
+    "producer_batch_seconds": 0.09067200803912101,
+    "producer_p95_seconds": 0.3466516779990343,
+    "gpu_update_seconds": 0.9836821027531235,
+    "producer_batches_per_second": 11.028762036113104,
+    "gpu_batches_per_second": 1.016588588123344,
+    "producer_headroom": 10.848795830447559,
     "sustainable": true,
-    "buffer_seconds": 60.32944278472757,
+    "buffer_seconds": 60.00460826794053,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 252290.11487268397,
+    "generation_bytes_per_second": 282064.2376143455,
     "upload_bytes_per_second": 2923200.839812715,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.135585638383766,
+    "estimated_full_shards_per_hour": 60.52441927263998,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -299,3 +299,4 @@
 {"step": 224900, "loss": 0.5770238377153873, "lr": 0.0003, "grad_norm": 0.13778783520921162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32918676145, "tokens_per_second": 149961.2548642165, "data_wait_s": 1.3821998436469585e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 225000, "loss": 0.5581230074167252, "lr": 0.0003, "grad_norm": 0.14314478292067695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32933375958, "tokens_per_second": 149900.43056667518, "data_wait_s": 1.671599966357462e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 225100, "loss": 0.5910075828433037, "lr": 0.0003, "grad_norm": 0.1438496628176133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32948012336, "tokens_per_second": 133571.83539380494, "data_wait_s": 1.038000118569471e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 225200, "loss": 0.6066942475736141, "lr": 0.0003, "grad_norm": 0.14503524494823936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32962651332, "tokens_per_second": 117636.08793781688, "data_wait_s": 1.7193000530824065e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0bf0296a222d34817cb0bb1872b70e0de6357005e04f900e14f9755f3fbc0493
-size 642381
+oid sha256:156bd0cb482b8738c50e06829cf0b2ad8234ac68529213bd961dee636af057a8
+size 644499