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": 76551.41430927317,
-      "required_pages_per_second_upper_estimate": 0.0052960044656684925
+      "required_token_positions_per_second": 75557.02234549109,
+      "required_pages_per_second_upper_estimate": 0.005227210122306807
     },
     {
       "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": 9568.926788659146,
-      "required_pages_per_second_upper_estimate": 0.00400630982449773
+      "required_token_positions_per_second": 9444.627793186386,
+      "required_pages_per_second_upper_estimate": 0.003954268430751473
     },
     {
       "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": 19137.853577318292,
-      "required_pages_per_second_upper_estimate": 0.0039859102010170195
+      "required_token_positions_per_second": 18889.255586372772,
+      "required_pages_per_second_upper_estimate": 0.003934133795472959
     },
     {
       "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": 28706.78036597744,
-      "required_pages_per_second_upper_estimate": 0.0017209276502247253
+      "required_token_positions_per_second": 28333.883379559156,
+      "required_pages_per_second_upper_estimate": 0.0016985730452696792
     },
     {
       "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": 28706.78036597744,
-      "required_pages_per_second_upper_estimate": 0.0022318259056723065
+      "required_token_positions_per_second": 28333.883379559156,
+      "required_pages_per_second_upper_estimate": 0.002202834805178787
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28706.78036597744,
-      "required_pages_per_second_upper_estimate": 0.3682197555954572
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28333.883379559156,
+      "required_pages_per_second_upper_estimate": 0.3634366334392729
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.46312074129667735,
-    "producer_p95_seconds": 0.37570127800427144,
-    "gpu_update_seconds": 0.963117411549493,
-    "producer_batches_per_second": 2.159264120194943,
-    "gpu_batches_per_second": 1.0382950074499941,
-    "producer_headroom": 2.079624870293847,
+    "producer_batch_seconds": 0.14732800978117666,
+    "producer_p95_seconds": 0.3724300760004553,
+    "gpu_update_seconds": 0.9757928212532289,
+    "producer_batches_per_second": 6.7875755702210325,
+    "gpu_batches_per_second": 1.024807703253731,
+    "producer_headroom": 6.623267515135475,
     "sustainable": true,
-    "buffer_seconds": 60.676396927618065,
+    "buffer_seconds": 60.49915491770019,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336032.59633999126,
-    "upload_bytes_per_second": 5261988.88497096,
+    "generation_bytes_per_second": 323196.7795565917,
+    "upload_bytes_per_second": 5262485.8609987255,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.10477273607066,
+    "estimated_full_shards_per_hour": 69.35050525687517,
     "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": 1.2377208289981354,
-      "bytes": 11067407,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.5320398900003056,
       "bytes": 7616229,
@@ -562,6 +556,12 @@
       "bytes": 1241882,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.234218167999643,
+      "bytes": 11088956,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -398,3 +398,4 @@
 {"step": 480000, "loss": 0.6134145259857178, "lr": 0.0003, "grad_norm": 0.15646532583825484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70256414402, "tokens_per_second": 152872.12864308245, "data_wait_s": 1.1411000741645694e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 480100, "loss": 0.5854529738426208, "lr": 0.0003, "grad_norm": 0.144068183056573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70271254957, "tokens_per_second": 134565.61914146633, "data_wait_s": 1.5833000361453742e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 480200, "loss": 0.6237391829490662, "lr": 0.0003, "grad_norm": 0.15932251469272685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70285977100, "tokens_per_second": 152124.1980707504, "data_wait_s": 8.499999239575118e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 480300, "loss": 0.5954119563102722, "lr": 0.0003, "grad_norm": 0.17127996462042347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70300565452, "tokens_per_second": 151434.82687737868, "data_wait_s": 1.6024998330976814e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e14c1e4c6198531eb42162ae98aee10fee8d36b5b74225fcc14e854c04eff6d8
-size 853629
+oid sha256:1c8111b144de68bd74406cfba94b9b1576de8a1450868d72d271bad29aed48df
+size 855747