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": 75252.1952446676,
-      "required_pages_per_second_upper_estimate": 0.006522669558862632
+      "required_token_positions_per_second": 74501.16203502286,
+      "required_pages_per_second_upper_estimate": 0.006457571903726893
     },
     {
       "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": 9406.52440558345,
-      "required_pages_per_second_upper_estimate": 0.0037524032254601285
+      "required_token_positions_per_second": 9312.645254377858,
+      "required_pages_per_second_upper_estimate": 0.0037149534284258248
     },
     {
       "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": 18813.0488111669,
-      "required_pages_per_second_upper_estimate": 0.0043520677441127986
+      "required_token_positions_per_second": 18625.290508755716,
+      "required_pages_per_second_upper_estimate": 0.004308633165283233
     },
     {
       "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": 28219.57321675035,
-      "required_pages_per_second_upper_estimate": 0.0017184239522418493
+      "required_token_positions_per_second": 27937.935763133573,
+      "required_pages_per_second_upper_estimate": 0.001701273708954109
     },
     {
       "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": 28219.57321675035,
-      "required_pages_per_second_upper_estimate": 0.002172420863003802
+      "required_token_positions_per_second": 27937.935763133573,
+      "required_pages_per_second_upper_estimate": 0.002150739632202013
     },
     {
       "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": 28219.57321675035,
-      "required_pages_per_second_upper_estimate": 0.3503926545159411
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27937.935763133573,
+      "required_pages_per_second_upper_estimate": 0.3468956599219436
     }
   ],
   "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": 0.06559814619532034,
-    "producer_p95_seconds": 0.16381389900016075,
-    "gpu_update_seconds": 0.9797455045701726,
-    "producer_batches_per_second": 15.244333231955544,
-    "gpu_batches_per_second": 1.0206732210919542,
-    "producer_headroom": 14.935566954178135,
+    "producer_batch_seconds": 0.08436884734378225,
+    "producer_p95_seconds": 0.177839275000224,
+    "gpu_update_seconds": 0.9896221479785859,
+    "producer_batches_per_second": 11.852716156299334,
+    "gpu_batches_per_second": 1.0104866812475974,
+    "producer_headroom": 11.729710421977435,
     "sustainable": true,
-    "buffer_seconds": 60.744221283350704,
+    "buffer_seconds": 60.36695102669374,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430368.9981679146,
-    "upload_bytes_per_second": 6130657.500902683,
+    "generation_bytes_per_second": 428661.8831870372,
+    "upload_bytes_per_second": 6146991.177622631,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.34716852930143,
+    "estimated_full_shards_per_hour": 91.98086139400803,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 10868715,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.1746170310007074,
+      "bytes": 7476573,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -43,3 +43,4 @@
 {"step": 44300, "loss": 0.6200189627707005, "lr": 0.0003, "grad_norm": 0.22031305739333554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6482772528, "tokens_per_second": 149493.1879755575, "data_wait_s": 1.0878000466618687e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 44400, "loss": 0.6678473874926567, "lr": 0.0003, "grad_norm": 0.14349906677865265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6497548429, "tokens_per_second": 151151.85950591083, "data_wait_s": 7.697000000916887e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 44500, "loss": 0.5941246263682842, "lr": 0.0003, "grad_norm": 0.14556532383493165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6512179341, "tokens_per_second": 148924.71859996233, "data_wait_s": 1.2671999684243929e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 44600, "loss": 0.6809761300683022, "lr": 0.0003, "grad_norm": 0.13601700008440684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6526582608, "tokens_per_second": 146852.09765430127, "data_wait_s": 9.774999853107147e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c8f10123807125c8fd1a0294a66f09cc50e777f437df2028d0960e60fd89a2e
-size 95475
+oid sha256:121b9a19329203a2de661fdcc93e4bc007591260c02a459581420b4555dd182e
+size 97593