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": 77196.18994033794,
-      "required_pages_per_second_upper_estimate": 0.005340611539911138
+      "required_token_positions_per_second": 77194.52433881958,
+      "required_pages_per_second_upper_estimate": 0.00534049630973337
     },
     {
       "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": 9649.523742542242,
-      "required_pages_per_second_upper_estimate": 0.0040400540860327985
+      "required_token_positions_per_second": 9649.315542352448,
+      "required_pages_per_second_upper_estimate": 0.004039966916961046
     },
     {
       "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": 19299.047485084484,
-      "required_pages_per_second_upper_estimate": 0.004019482641035504
+      "required_token_positions_per_second": 19298.631084704895,
+      "required_pages_per_second_upper_estimate": 0.004019395915817652
     },
     {
       "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": 28948.571227626722,
-      "required_pages_per_second_upper_estimate": 0.001735422643187331
+      "required_token_positions_per_second": 28947.946627057343,
+      "required_pages_per_second_upper_estimate": 0.0017353851993369065
     },
     {
       "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": 28948.571227626722,
-      "required_pages_per_second_upper_estimate": 0.002250624081639935
+      "required_token_positions_per_second": 28947.946627057343,
+      "required_pages_per_second_upper_estimate": 0.002250575521692993
     },
     {
       "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": 28948.571227626722,
-      "required_pages_per_second_upper_estimate": 0.3713211891538939
+      "required_token_positions_per_second": 28947.946627057343,
+      "required_pages_per_second_upper_estimate": 0.37131317744843373
     }
   ],
   "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": 0.11460079114050359,
-    "producer_p95_seconds": 0.33192167000379413,
-    "gpu_update_seconds": 0.9550730425553597,
-    "producer_batches_per_second": 8.725943250897576,
-    "gpu_batches_per_second": 1.0470403366473786,
-    "producer_headroom": 8.333913169800155,
+    "producer_batch_seconds": 0.11925656821881603,
+    "producer_p95_seconds": 0.35343281099630985,
+    "gpu_update_seconds": 0.9550936498603912,
+    "producer_batches_per_second": 8.385282378452866,
+    "gpu_batches_per_second": 1.0470177454809513,
+    "producer_headroom": 8.008729951946568,
     "sustainable": true,
-    "buffer_seconds": 60.16960168098766,
+    "buffer_seconds": 60.170899941204645,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348491.63000080624,
-    "upload_bytes_per_second": 4961585.028332936,
+    "generation_bytes_per_second": 331112.6281430495,
+    "upload_bytes_per_second": 4948994.516259112,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.77819132822171,
+    "estimated_full_shards_per_hour": 71.0490620920049,
     "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": 1.2240972300060093,
-      "bytes": 7561561,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 0.9214401089993771,
       "bytes": 1231310,
@@ -562,6 +556,12 @@
       "bytes": 1097506,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.100044867998804,
+      "bytes": 5862761,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -432,3 +432,4 @@
 {"step": 928200, "loss": 0.5369181036949158, "lr": 3e-05, "grad_norm": 0.32134237412797756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135862068058, "tokens_per_second": 151836.46588166346, "data_wait_s": 1.745599729474634e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 928300, "loss": 0.5516844391822815, "lr": 3e-05, "grad_norm": 0.3289543805894701, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135876730454, "tokens_per_second": 153243.61821007414, "data_wait_s": 1.0798998118843883e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 928400, "loss": 0.584831953048706, "lr": 3e-05, "grad_norm": 0.34174505322392906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135891380783, "tokens_per_second": 152270.48114996063, "data_wait_s": 1.6571000742260367e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 928500, "loss": 0.5752297043800354, "lr": 3e-05, "grad_norm": 0.35337629774470697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135906124229, "tokens_per_second": 153442.967841364, "data_wait_s": 1.0928997653536499e-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 sha256:467153966db97f5f323c8c26398b5bafbfcccb474583ae92aa1e78543fffd5e7
-size 925641
+oid sha256:1b1cd61737230a044d3fece875c8147b36a8ed8065e41a5f752321a9c15b4b6f
+size 927759