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": 75525.95179162378,
-      "required_pages_per_second_upper_estimate": 0.00522506058929665
+      "required_token_positions_per_second": 77008.88924755574,
+      "required_pages_per_second_upper_estimate": 0.0053276536433869825
     },
     {
       "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": 9440.743973952973,
-      "required_pages_per_second_upper_estimate": 0.003952642356741811
+      "required_token_positions_per_second": 9626.111155944467,
+      "required_pages_per_second_upper_estimate": 0.004030251724934714
     },
     {
       "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": 18881.487947905945,
-      "required_pages_per_second_upper_estimate": 0.003932516001226721
+      "required_token_positions_per_second": 19252.222311888934,
+      "required_pages_per_second_upper_estimate": 0.004009730192321729
     },
     {
       "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": 28322.23192185892,
-      "required_pages_per_second_upper_estimate": 0.0016978745581712957
+      "required_token_positions_per_second": 28878.333467833403,
+      "required_pages_per_second_upper_estimate": 0.0017312119967345708
     },
     {
       "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": 28322.23192185892,
-      "required_pages_per_second_upper_estimate": 0.0022019289555919395
+      "required_token_positions_per_second": 28878.333467833403,
+      "required_pages_per_second_upper_estimate": 0.0022451634047592596
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28322.23192185892,
-      "required_pages_per_second_upper_estimate": 0.36328718105025487
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28878.333467833403,
+      "required_pages_per_second_upper_estimate": 0.370420254586696
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10576167001556769,
-    "producer_p95_seconds": 0.3213382969988743,
-    "gpu_update_seconds": 0.9761942517906383,
-    "producer_batches_per_second": 9.455221346758274,
-    "gpu_batches_per_second": 1.0243862818959388,
-    "producer_headroom": 9.230132728113565,
+    "producer_batch_seconds": 0.10245465230485706,
+    "producer_p95_seconds": 0.3182921309999074,
+    "gpu_update_seconds": 0.9573959671459633,
+    "producer_batches_per_second": 9.760415730312259,
+    "gpu_batches_per_second": 1.044499908414113,
+    "producer_headroom": 9.344582657868978,
     "sustainable": true,
-    "buffer_seconds": 60.52404361101958,
+    "buffer_seconds": 60.31594593019569,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 548026.0667861896,
+    "generation_bytes_per_second": 526637.8420752296,
     "upload_bytes_per_second": 3837858.195923135,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 117.59363653840317,
+    "estimated_full_shards_per_hour": 113.00422140782038,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -341,3 +341,4 @@
 {"step": 834100, "loss": 0.5463454127311707, "lr": 3e-05, "grad_norm": 0.29014100913848845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122088582622, "tokens_per_second": 152620.16902439186, "data_wait_s": 1.592100306879729e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 834200, "loss": 0.5487772822380066, "lr": 3e-05, "grad_norm": 0.2698189661515686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122103315124, "tokens_per_second": 152552.07153830057, "data_wait_s": 1.068100391421467e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 834300, "loss": 0.5587754845619202, "lr": 3e-05, "grad_norm": 0.2948067504570078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122118046312, "tokens_per_second": 153036.59667053417, "data_wait_s": 1.3212003977969289e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 834400, "loss": 0.5076045393943787, "lr": 3e-05, "grad_norm": 0.2948557973666099, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122132533332, "tokens_per_second": 150243.17243182645, "data_wait_s": 1.3503005902748555e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39cd11e5edf6177d092725db3874b6d5c4e2b19c0ff6d9c22c1d83e713cb25c7
-size 731337
+oid sha256:5834d3587439062fad86f216481884b911761223fac6df5245536043abd72546
+size 733455