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": 74966.59018514432,
-      "required_pages_per_second_upper_estimate": 0.0064979140361625595
+      "required_token_positions_per_second": 75867.11059377519,
+      "required_pages_per_second_upper_estimate": 0.0065759688628345735
     },
     {
       "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": 9370.82377314304,
-      "required_pages_per_second_upper_estimate": 0.003738161709407627
+      "required_token_positions_per_second": 9483.388824221898,
+      "required_pages_per_second_upper_estimate": 0.0037830655912804764
     },
     {
       "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": 18741.64754628608,
-      "required_pages_per_second_upper_estimate": 0.004335550318101964
+      "required_token_positions_per_second": 18966.777648443796,
+      "required_pages_per_second_upper_estimate": 0.004387630205081678
     },
     {
       "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": 28112.47131942912,
-      "required_pages_per_second_upper_estimate": 0.0017119020086152144
+      "required_token_positions_per_second": 28450.166472665693,
+      "required_pages_per_second_upper_estimate": 0.0017324658716977807
     },
     {
       "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": 28112.47131942912,
-      "required_pages_per_second_upper_estimate": 0.002164175862470987
+      "required_token_positions_per_second": 28450.166472665693,
+      "required_pages_per_second_upper_estimate": 0.002190172570167152
     },
     {
       "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": 28112.47131942912,
-      "required_pages_per_second_upper_estimate": 0.34906280739820356
+      "required_token_positions_per_second": 28450.166472665693,
+      "required_pages_per_second_upper_estimate": 0.3532558510084271
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.333108570421885,
-    "producer_p95_seconds": 0.20775366599991685,
-    "gpu_update_seconds": 0.9834781042850503,
-    "producer_batches_per_second": 3.0020242311192744,
-    "gpu_batches_per_second": 1.0167994545511112,
-    "producer_headroom": 2.95242509983897,
+    "producer_batch_seconds": 0.22093081212499754,
+    "producer_p95_seconds": 0.19389570500061382,
+    "gpu_update_seconds": 0.9718045068932585,
+    "producer_batches_per_second": 4.526303915608762,
+    "gpu_batches_per_second": 1.0290135442949109,
+    "producer_headroom": 4.3986825447571976,
     "sustainable": true,
-    "buffer_seconds": 60.97564246567312,
+    "buffer_seconds": 60.25187942738203,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 462287.0798362098,
-    "upload_bytes_per_second": 5528470.75859466,
+    "generation_bytes_per_second": 445975.07795824006,
+    "upload_bytes_per_second": 5673641.739379737,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.19604584040376,
+    "estimated_full_shards_per_hour": 95.69586996136094,
     "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": [
@@ -388,6 +388,12 @@
       "bytes": 1120534,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.3186273349992916,
+      "bytes": 13209471,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -67,3 +67,4 @@
 {"step": 361700, "loss": 0.6090059615671635, "lr": 0.0003, "grad_norm": 0.16071528306354366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52943041541, "tokens_per_second": 151162.04550983655, "data_wait_s": 1.5300000086426735e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 361800, "loss": 0.5769853852689266, "lr": 0.0003, "grad_norm": 0.17652536240169736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52957703659, "tokens_per_second": 149332.9835808498, "data_wait_s": 1.167800019175047e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 361900, "loss": 0.6227873042225838, "lr": 0.0003, "grad_norm": 0.15211010385266396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52972141502, "tokens_per_second": 147651.26659480113, "data_wait_s": 1.0863999705179594e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 362000, "loss": 0.5736682824790478, "lr": 0.0003, "grad_norm": 0.15145500118993704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52986652671, "tokens_per_second": 146843.54463762057, "data_wait_s": 8.873999831848778e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f83f0b55a850ca110e9bb852135b4e5e21fb566003f7370d20f82e78f12dadf2
-size 147090
+oid sha256:49931472741e96ec6e0aafc19ef8c2e06db3f60502c1a61a2c1b830468dbebb6
+size 149208