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": 75451.6732177696,
-      "required_pages_per_second_upper_estimate": 0.0065399598039988125
+      "required_token_positions_per_second": 76195.40407516385,
+      "required_pages_per_second_upper_estimate": 0.006604424509749122
     },
     {
       "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": 9431.4591522212,
-      "required_pages_per_second_upper_estimate": 0.0037623500687016115
+      "required_token_positions_per_second": 9524.425509395482,
+      "required_pages_per_second_upper_estimate": 0.0037994357385493383
     },
     {
       "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": 18862.9183044424,
-      "required_pages_per_second_upper_estimate": 0.00436360417370898
+      "required_token_positions_per_second": 19048.851018790963,
+      "required_pages_per_second_upper_estimate": 0.0044066164348695105
     },
     {
       "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": 28294.3774566636,
-      "required_pages_per_second_upper_estimate": 0.001722979137985051
+      "required_token_positions_per_second": 28573.276528186445,
+      "required_pages_per_second_upper_estimate": 0.0017399626281704515
     },
     {
       "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": 28294.3774566636,
-      "required_pages_per_second_upper_estimate": 0.002178179500463713
+      "required_token_positions_per_second": 28573.276528186445,
+      "required_pages_per_second_upper_estimate": 0.002199649923031585
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28294.3774566636,
-      "required_pages_per_second_upper_estimate": 0.35132147282197745
+      "required_token_positions_per_second": 28573.276528186445,
+      "required_pages_per_second_upper_estimate": 0.3547844658751437
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1053994430625167,
-    "producer_p95_seconds": 0.18592662299943186,
-    "gpu_update_seconds": 0.9771552684750316,
-    "producer_batches_per_second": 9.487716167597387,
-    "gpu_batches_per_second": 1.0233788142601128,
-    "producer_headroom": 9.270971838963524,
+    "producer_batch_seconds": 0.0722688109531262,
+    "producer_p95_seconds": 0.19211949200052914,
+    "gpu_update_seconds": 0.9676174159699993,
+    "producer_batches_per_second": 13.837227800089355,
+    "gpu_batches_per_second": 1.033466309613225,
+    "producer_headroom": 13.3891426081107,
     "sustainable": true,
-    "buffer_seconds": 60.583626645451965,
+    "buffer_seconds": 60.95989720610996,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405462.78213829803,
+    "generation_bytes_per_second": 421913.588886585,
     "upload_bytes_per_second": 6003453.127983018,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.00287435638147,
+    "estimated_full_shards_per_hour": 90.53283452938236,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -54,3 +54,4 @@
 {"step": 45400, "loss": 0.6239417102187872, "lr": 0.0003, "grad_norm": 0.1296677709303981, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6643543499, "tokens_per_second": 146850.14593881185, "data_wait_s": 1.4342000213218853e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 45500, "loss": 0.6155837811529636, "lr": 0.0003, "grad_norm": 0.13159858412503486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6658076638, "tokens_per_second": 147011.46590095738, "data_wait_s": 1.3998000213177875e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 45600, "loss": 0.6086126398295164, "lr": 0.0003, "grad_norm": 0.11961747347757953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6672697519, "tokens_per_second": 149299.29169207055, "data_wait_s": 1.3101999684295151e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 45700, "loss": 0.6558852419257164, "lr": 0.0003, "grad_norm": 0.1374799079946057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6687336151, "tokens_per_second": 148993.07911957163, "data_wait_s": 9.60300076258136e-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:f891cf5598e8966d6142c49cc51f7bf20195d1f372464c3d8f1da46584b90835
-size 119556
+oid sha256:344214c996194790c2a14dd51247029454afe93e0b7be6e6fb3ce71b6f035ccb
+size 121674