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": 77724.64051264,
-      "required_pages_per_second_upper_estimate": 0.005377170976677257
+      "required_token_positions_per_second": 77004.71756162532,
+      "required_pages_per_second_upper_estimate": 0.005327365036474671
     },
     {
       "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": 9715.58006408,
-      "required_pages_per_second_upper_estimate": 0.004067710488447806
+      "required_token_positions_per_second": 9625.589695203165,
+      "required_pages_per_second_upper_estimate": 0.00403003340021167
     },
     {
       "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": 19431.16012816,
-      "required_pages_per_second_upper_estimate": 0.004046998220543444
+      "required_token_positions_per_second": 19251.17939040633,
+      "required_pages_per_second_upper_estimate": 0.004009512979280592
     },
     {
       "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": 29146.74019224,
-      "required_pages_per_second_upper_estimate": 0.0017473025700294083
+      "required_token_positions_per_second": 28876.769085609496,
+      "required_pages_per_second_upper_estimate": 0.001731118214408921
     },
     {
       "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": 29146.74019224,
-      "required_pages_per_second_upper_estimate": 0.002266030846985461
+      "required_token_positions_per_second": 28876.769085609496,
+      "required_pages_per_second_upper_estimate": 0.002245041780922341
     },
     {
       "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": 29146.74019224,
-      "required_pages_per_second_upper_estimate": 0.3738630878546966
+      "required_token_positions_per_second": 28876.769085609496,
+      "required_pages_per_second_upper_estimate": 0.37040018837123045
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.550541744593545,
-    "producer_p95_seconds": 0.3217196959958528,
-    "gpu_update_seconds": 0.9485794918280768,
-    "producer_batches_per_second": 1.8163926892379103,
-    "gpu_batches_per_second": 1.0542079062586807,
-    "producer_headroom": 1.7229928541175308,
+    "producer_batch_seconds": 0.42237205035132774,
+    "producer_p95_seconds": 0.3190965769972536,
+    "gpu_update_seconds": 0.9574478335174331,
+    "producer_batches_per_second": 2.3675809021174654,
+    "gpu_batches_per_second": 1.0444433263024269,
+    "producer_headroom": 2.266835205409617,
     "sustainable": true,
-    "buffer_seconds": 60.70908747699691,
+    "buffer_seconds": 60.319213511598285,
     "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": 259221.93518007378,
-    "upload_bytes_per_second": 5131968.739206977,
+    "generation_bytes_per_second": 246032.14424401076,
+    "upload_bytes_per_second": 5084831.193985445,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.622992911831474,
+    "estimated_full_shards_per_hour": 52.79277081957095,
     "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": 1.1933262150014343,
-      "bytes": 11280110,
-      "pages": 36,
-      "shards": 5
-    },
     {
       "seconds": 1.0185086079982284,
       "bytes": 1287749,
@@ -562,6 +556,12 @@
       "bytes": 11134689,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 1.0753672879945952,
+      "bytes": 6680589,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -396,3 +396,4 @@
 {"step": 924600, "loss": 0.4500882625579834, "lr": 3e-05, "grad_norm": 0.3676246642681876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135335218133, "tokens_per_second": 153495.27343306196, "data_wait_s": 1.347699435427785e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 924700, "loss": 0.5373215079307556, "lr": 3e-05, "grad_norm": 0.3539644410212644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135349890405, "tokens_per_second": 153375.5387203123, "data_wait_s": 1.5151003026403487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 924800, "loss": 0.5135881900787354, "lr": 3e-05, "grad_norm": 0.34735354490061177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135364418314, "tokens_per_second": 140280.27799993314, "data_wait_s": 1.615499786566943e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 924900, "loss": 0.5153971910476685, "lr": 3e-05, "grad_norm": 0.3553751987668995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135379093364, "tokens_per_second": 152545.82118496113, "data_wait_s": 1.914300082717091e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7d180be3ba5abb74f12ae1650c20739e233b55461aae1dea151d5082d8add61b
-size 848610
+oid sha256:97b0f8aa0c0673b1da11af7ec508bdcf4c66781b71a06ed3ef4d8c0fe7c9b6c8
+size 850728