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": 74063.19918959186,
-      "required_pages_per_second_upper_estimate": 0.0064196103942916185
+      "required_token_positions_per_second": 74544.01279364803,
+      "required_pages_per_second_upper_estimate": 0.006461286098880268
     },
     {
       "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": 9257.899898698983,
-      "required_pages_per_second_upper_estimate": 0.003693114687529513
+      "required_token_positions_per_second": 9318.001599206003,
+      "required_pages_per_second_upper_estimate": 0.003717090154462264
     },
     {
       "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": 18515.799797397965,
-      "required_pages_per_second_upper_estimate": 0.004283304416181324
+      "required_token_positions_per_second": 18636.003198412007,
+      "required_pages_per_second_upper_estimate": 0.004311111357498318
     },
     {
       "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": 27773.69969609695,
-      "required_pages_per_second_upper_estimate": 0.0016912725941516285
+      "required_token_positions_per_second": 27954.004797618007,
+      "required_pages_per_second_upper_estimate": 0.0017022522288465027
     },
     {
       "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": 27773.69969609695,
-      "required_pages_per_second_upper_estimate": 0.0021380962851270027
+      "required_token_positions_per_second": 27954.004797618007,
+      "required_pages_per_second_upper_estimate": 0.002151976670958559
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27773.69969609695,
-      "required_pages_per_second_upper_estimate": 0.3448563976321064
+      "required_token_positions_per_second": 27954.004797618007,
+      "required_pages_per_second_upper_estimate": 0.3470951835506414
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3553039287578237,
-    "producer_p95_seconds": 0.18037243600019792,
-    "gpu_update_seconds": 0.9954741464957002,
-    "producer_batches_per_second": 2.814491816896298,
-    "gpu_batches_per_second": 1.0045464299803584,
-    "producer_headroom": 2.801753839243975,
+    "producer_batch_seconds": 0.049605360054702174,
+    "producer_p95_seconds": 0.17073626100000183,
+    "gpu_update_seconds": 0.9890532751985475,
+    "producer_batches_per_second": 20.159111815683886,
+    "gpu_batches_per_second": 1.0110678818582903,
+    "producer_headroom": 19.93843556639588,
     "sustainable": true,
-    "buffer_seconds": 60.72392293623771,
+    "buffer_seconds": 60.332249787111394,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441987.65919450193,
-    "upload_bytes_per_second": 3377826.7471939926,
+    "generation_bytes_per_second": 442485.9707016218,
+    "upload_bytes_per_second": 3565333.9420681917,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.84026271702092,
+    "estimated_full_shards_per_hour": 94.94718876635065,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 3757694,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.5062500680000994,
+      "bytes": 15040116,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -28,3 +28,4 @@
 {"step": 2800, "loss": 0.8120828606188297, "lr": 0.0003, "grad_norm": 0.220045838165631, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 409819441, "tokens_per_second": 147645.89649371261, "data_wait_s": 1.4212999758456135e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 2900, "loss": 0.8301848135888577, "lr": 0.0003, "grad_norm": 0.18329583427996232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 424456851, "tokens_per_second": 148737.95268198178, "data_wait_s": 1.147700004366925e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 3000, "loss": 0.8225421886891127, "lr": 0.0003, "grad_norm": 0.20396286009016903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 439054516, "tokens_per_second": 148361.7991957803, "data_wait_s": 1.125900007536984e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 3100, "loss": 0.8211501948535442, "lr": 0.0003, "grad_norm": 0.21141356178096296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 453545230, "tokens_per_second": 147444.06545955865, "data_wait_s": 1.2337000043771695e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e5aa4d34bb8f1cf03577be70820dd20db1c625e7947548c9cd3a3365d6918fd9
-size 62616
+oid sha256:177a74da4aaef2d2412a1f62bf63080ef6c190885c7c28eff6d7d79d57b39844
+size 64699