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": 76056.38417927203,
-      "required_pages_per_second_upper_estimate": 0.006592374617515924
+      "required_token_positions_per_second": 75008.29072648216,
+      "required_pages_per_second_upper_estimate": 0.0065015285333966675
     },
     {
       "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": 9507.048022409004,
-      "required_pages_per_second_upper_estimate": 0.003792503599173849
+      "required_token_positions_per_second": 9376.03634081027,
+      "required_pages_per_second_upper_estimate": 0.003740241080584917
     },
     {
       "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": 19014.09604481801,
-      "required_pages_per_second_upper_estimate": 0.0043985764833075185
+      "required_token_positions_per_second": 18752.07268162054,
+      "required_pages_per_second_upper_estimate": 0.0043379619896854755
     },
     {
       "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": 28521.144067227015,
-      "required_pages_per_second_upper_estimate": 0.0017367880348159064
+      "required_token_positions_per_second": 28128.109022430806,
+      "required_pages_per_second_upper_estimate": 0.0017128542626833301
     },
     {
       "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": 28521.144067227015,
-      "required_pages_per_second_upper_estimate": 0.0021956366218750397
+      "required_token_positions_per_second": 28128.109022430806,
+      "required_pages_per_second_upper_estimate": 0.002165379696135989
     },
     {
       "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": 28521.144067227015,
-      "required_pages_per_second_upper_estimate": 0.35413715518615063
+      "required_token_positions_per_second": 28128.109022430806,
+      "required_pages_per_second_upper_estimate": 0.34925697533345923
     }
   ],
   "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": 0.16100460467157518,
-    "producer_p95_seconds": 0.4148923309985548,
-    "gpu_update_seconds": 0.9693860784417017,
-    "producer_batches_per_second": 6.211002486791277,
-    "gpu_batches_per_second": 1.0315807315981993,
-    "producer_headroom": 6.020859343862254,
+    "producer_batch_seconds": 0.10651884664042655,
+    "producer_p95_seconds": 0.4051830219978001,
+    "gpu_update_seconds": 0.9829313438010375,
+    "producer_batches_per_second": 9.388010023950777,
+    "gpu_batches_per_second": 1.017365054341392,
+    "producer_headroom": 9.227769308459548,
     "sustainable": true,
-    "buffer_seconds": 60.101936863385504,
+    "buffer_seconds": 60.941743315664326,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280402.8991853839,
-    "upload_bytes_per_second": 5839661.826132428,
+    "generation_bytes_per_second": 280553.04828175006,
+    "upload_bytes_per_second": 5784040.446851314,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.16793471976412,
+    "estimated_full_shards_per_hour": 60.20015322055222,
     "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": 1.3008259939997515,
-      "bytes": 12099101,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.8470429570006672,
       "bytes": 932725,
@@ -562,6 +556,12 @@
       "bytes": 2315430,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 0.974858529996709,
+      "bytes": 6294959,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -391,3 +391,4 @@
 {"step": 154100, "loss": 0.6051191464066505, "lr": 0.0003, "grad_norm": 0.14220059304235802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22552412707, "tokens_per_second": 149023.50628410318, "data_wait_s": 1.20009935926646e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 154200, "loss": 0.6281894072890282, "lr": 0.0003, "grad_norm": 0.1405169190934155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22567129027, "tokens_per_second": 150378.26612644913, "data_wait_s": 1.2297998182475567e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 154300, "loss": 0.6565862763673067, "lr": 0.0003, "grad_norm": 0.15910598349115007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22581817392, "tokens_per_second": 149324.08239671055, "data_wait_s": 1.1727999662980437e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 154400, "loss": 0.6272453535348177, "lr": 0.0003, "grad_norm": 0.13534144886402907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22596594148, "tokens_per_second": 149704.79556514946, "data_wait_s": 1.288699422730133e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2fcc30118b4077e18352d4033d23062062e0e1ab0f8c95d0d9782fde4de47dc3
-size 838020
+oid sha256:dc161ec581aa173d588135aeead1fbf8e7ef3a8b1479e8b5367d158683e7ff0c
+size 840138