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": 76359.26031322179,
-      "required_pages_per_second_upper_estimate": 0.00528271080636296
+      "required_token_positions_per_second": 77223.72760759271,
+      "required_pages_per_second_upper_estimate": 0.005342516659628014
     },
     {
       "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": 9544.907539152724,
-      "required_pages_per_second_upper_estimate": 0.003996253466308357
+      "required_token_positions_per_second": 9652.965950949088,
+      "required_pages_per_second_upper_estimate": 0.00404149526681126
     },
     {
       "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": 19089.815078305448,
-      "required_pages_per_second_upper_estimate": 0.003975905048533055
+      "required_token_positions_per_second": 19305.931901898177,
+      "required_pages_per_second_upper_estimate": 0.004020916483503516
     },
     {
       "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": 28634.72261745817,
-      "required_pages_per_second_upper_estimate": 0.0017166078982268063
+      "required_token_positions_per_second": 28958.897852847265,
+      "required_pages_per_second_upper_estimate": 0.0017360417086014588
     },
     {
       "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": 28634.72261745817,
-      "required_pages_per_second_upper_estimate": 0.002226223732673473
+      "required_token_positions_per_second": 28958.897852847265,
+      "required_pages_per_second_upper_estimate": 0.002251426931328808
     },
     {
       "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": 28634.72261745817,
-      "required_pages_per_second_upper_estimate": 0.3672954761670344
+      "required_token_positions_per_second": 28958.897852847265,
+      "required_pages_per_second_upper_estimate": 0.3714536480143567
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14222168188268824,
-    "producer_p95_seconds": 0.32549434400425525,
-    "gpu_update_seconds": 0.96554104502285,
-    "producer_batches_per_second": 7.031276713664878,
-    "gpu_batches_per_second": 1.0356887520782034,
-    "producer_headroom": 6.788986265956817,
+    "producer_batch_seconds": 0.16101039687532648,
+    "producer_p95_seconds": 0.33448731699900236,
+    "gpu_update_seconds": 0.9547324673919392,
+    "producer_batches_per_second": 6.2107790515808725,
+    "gpu_batches_per_second": 1.0474138401637465,
+    "producer_headroom": 5.929632408341974,
     "sustainable": true,
-    "buffer_seconds": 60.829085836439546,
+    "buffer_seconds": 60.148145445692165,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325082.9739961289,
-    "upload_bytes_per_second": 2979475.1226245803,
+    "generation_bytes_per_second": 327384.2572764032,
+    "upload_bytes_per_second": 3002904.618608022,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.75523867524052,
+    "estimated_full_shards_per_hour": 70.24904049605438,
     "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": 1.0848596139985602,
-      "bytes": 2384316,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 2.864481454998895,
       "bytes": 16026880,
@@ -562,6 +556,12 @@
       "bytes": 2681709,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 2.4406843360047787,
+      "bytes": 9815983,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -411,3 +411,4 @@
 {"step": 883600, "loss": 0.5222635269165039, "lr": 3e-05, "grad_norm": 0.3101059288032435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129334124805, "tokens_per_second": 151940.3420529686, "data_wait_s": 9.590003173798323e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 883700, "loss": 0.5343281030654907, "lr": 3e-05, "grad_norm": 0.3070176355470963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129348785742, "tokens_per_second": 152509.59901306962, "data_wait_s": 1.186799636343494e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 883800, "loss": 0.5805310010910034, "lr": 3e-05, "grad_norm": 0.32932567289974446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129363442301, "tokens_per_second": 152204.26088067185, "data_wait_s": 1.1217998689971864e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 883900, "loss": 0.5239691138267517, "lr": 3e-05, "grad_norm": 0.3012883191103843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129378229716, "tokens_per_second": 152706.0794318554, "data_wait_s": 1.2616001185961068e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:70dd82f308239dc188099230bc7e7721497f6404167599af17a30c53e82db233
-size 881163
+oid sha256:8c5e813478b527ccbed8bd3c46c5c731a6ce160b8a35703260820bdb0e6f6cd8
+size 883281