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": 76111.85832149869,
-      "required_pages_per_second_upper_estimate": 0.006597182975566202
+      "required_token_positions_per_second": 75929.9801598137,
+      "required_pages_per_second_upper_estimate": 0.006581418237477333
     },
     {
       "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": 9513.982290187336,
-      "required_pages_per_second_upper_estimate": 0.0037952697822671677
+      "required_token_positions_per_second": 9491.247519976712,
+      "required_pages_per_second_upper_estimate": 0.003786200542515044
     },
     {
       "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": 19027.96458037467,
-      "required_pages_per_second_upper_estimate": 0.004401784724930663
+      "required_token_positions_per_second": 18982.495039953425,
+      "required_pages_per_second_upper_estimate": 0.004391266146990787
     },
     {
       "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": 28541.94687056201,
-      "required_pages_per_second_upper_estimate": 0.0017380548163951335
+      "required_token_positions_per_second": 28473.742559930135,
+      "required_pages_per_second_upper_estimate": 0.0017339015317180148
     },
     {
       "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": 28541.94687056201,
-      "required_pages_per_second_upper_estimate": 0.0021972380792616127
+      "required_token_positions_per_second": 28473.742559930135,
+      "required_pages_per_second_upper_estimate": 0.0021919875226275557
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28541.94687056201,
-      "required_pages_per_second_upper_estimate": 0.3543954563810672
+      "required_token_positions_per_second": 28473.742559930135,
+      "required_pages_per_second_upper_estimate": 0.35354858710816317
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.39223933092984,
+    "producer_batch_seconds": 0.1148675779062387,
     "producer_p95_seconds": 0.29193562000000384,
-    "gpu_update_seconds": 0.9686795412164396,
-    "producer_batches_per_second": 2.549463863374962,
-    "gpu_batches_per_second": 1.0323331478067856,
-    "producer_headroom": 2.4696134855219496,
+    "gpu_update_seconds": 0.9709998586173857,
+    "producer_batches_per_second": 8.705676729914646,
+    "gpu_batches_per_second": 1.029866267358584,
+    "producer_headroom": 8.453210873915786,
     "sustainable": true,
-    "buffer_seconds": 60.05813155541926,
+    "buffer_seconds": 60.20199123427791,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322176.1943315808,
-    "upload_bytes_per_second": 5182151.64904447,
+    "generation_bytes_per_second": 322418.9202920593,
+    "upload_bytes_per_second": 5133938.350760595,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.13151142559593,
+    "estimated_full_shards_per_hour": 69.18359476634345,
     "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": 1.5922990050003136,
-      "bytes": 11693360,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 0.8287107050000486,
       "bytes": 2337614,
@@ -562,6 +556,12 @@
       "bytes": 6106174,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0782650920009473,
+      "bytes": 5102757,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -171,3 +171,4 @@
 {"step": 172100, "loss": 0.5721213407814503, "lr": 0.0003, "grad_norm": 0.14904676025056426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25187867206, "tokens_per_second": 149166.60639444797, "data_wait_s": 9.974999557016417e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 172200, "loss": 0.6270482540130615, "lr": 0.0003, "grad_norm": 0.13754360075202146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25202419943, "tokens_per_second": 140488.59696334766, "data_wait_s": 1.0435996955493465e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 172300, "loss": 0.6193122211843729, "lr": 0.0003, "grad_norm": 0.138339238126607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25217023351, "tokens_per_second": 149087.81708934403, "data_wait_s": 1.4073000784264877e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 172400, "loss": 0.5952528994530439, "lr": 0.0003, "grad_norm": 0.1611058560597149, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25231503329, "tokens_per_second": 147181.13615719057, "data_wait_s": 1.119299849960953e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2a25fa62dbafe6293e19fce617c1b8d128f8bb30379adbbc98df347e47e6ee21
-size 368928
+oid sha256:03dd1d791c799fbb89db83b69dc32df369d167d1544d80e3874e49431514ecd9
+size 371046