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": 76546.53393498673,
-      "required_pages_per_second_upper_estimate": 0.005295666830051328
+      "required_token_positions_per_second": 77355.60304842793,
+      "required_pages_per_second_upper_estimate": 0.005351640108618176
     },
     {
       "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": 9568.316741873341,
-      "required_pages_per_second_upper_estimate": 0.004006054410647739
+      "required_token_positions_per_second": 9669.450381053492,
+      "required_pages_per_second_upper_estimate": 0.0040483969534619284
     },
     {
       "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": 19136.633483746682,
-      "required_pages_per_second_upper_estimate": 0.003985656087702084
+      "required_token_positions_per_second": 19338.900762106983,
+      "required_pages_per_second_upper_estimate": 0.004027783027637699
     },
     {
       "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": 28704.950225620025,
-      "required_pages_per_second_upper_estimate": 0.0017208179361047607
+      "required_token_positions_per_second": 29008.351143160475,
+      "required_pages_per_second_upper_estimate": 0.0017390063578449336
     },
     {
       "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": 28704.950225620025,
-      "required_pages_per_second_upper_estimate": 0.0022316836203094543
+      "required_token_positions_per_second": 29008.351143160475,
+      "required_pages_per_second_upper_estimate": 0.0022552717071286244
     },
     {
       "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": 28704.950225620025,
-      "required_pages_per_second_upper_estimate": 0.3681962805200039
+      "required_token_positions_per_second": 29008.351143160475,
+      "required_pages_per_second_upper_estimate": 0.37208798172368845
     }
   ],
   "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": 0.2865978292658724,
-    "producer_p95_seconds": 0.4934889669966651,
-    "gpu_update_seconds": 0.9631788169875778,
-    "producer_batches_per_second": 3.4892099586431806,
-    "gpu_batches_per_second": 1.0382288131372983,
-    "producer_headroom": 3.360733120187214,
+    "producer_batch_seconds": 0.11286816035902802,
+    "producer_p95_seconds": 0.3355309879989363,
+    "gpu_update_seconds": 0.9531048443102835,
+    "producer_batches_per_second": 8.859894560335258,
+    "gpu_batches_per_second": 1.0492025153052833,
+    "producer_headroom": 8.444408425533863,
     "sustainable": true,
-    "buffer_seconds": 60.6802654702174,
+    "buffer_seconds": 60.045605191547864,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349135.0638823028,
-    "upload_bytes_per_second": 4995073.260094661,
+    "generation_bytes_per_second": 334736.7877699698,
+    "upload_bytes_per_second": 5057899.1768841725,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.91625726081669,
+    "estimated_full_shards_per_hour": 71.82672238182373,
     "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": 1.0402027899981476,
-      "bytes": 5156086,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.5196284359990386,
       "bytes": 12048583,
@@ -562,6 +556,12 @@
       "bytes": 6028853,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.3323977969994303,
+      "bytes": 12061980,
+      "pages": 38,
+      "shards": 5
     }
   ],
   "cache_key": {

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

@@ -430,3 +430,4 @@
 {"step": 928000, "loss": 0.5243933200836182, "lr": 3e-05, "grad_norm": 0.33932875903643783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135833040179, "tokens_per_second": 153850.70217473313, "data_wait_s": 1.2272997992113233e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 928100, "loss": 0.55768883228302, "lr": 3e-05, "grad_norm": 0.35417443810619026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135847590762, "tokens_per_second": 151379.7995990841, "data_wait_s": 1.7454003682360053e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:89472952228dda0b556542a5af20396b00fbc8d4d9d84e0313352b01612aac80
-size 921405
+oid sha256:c92589dfd9aca43dd845dd385aa351ab8992613d78c145de0b2f6bd7f3419d97
+size 923523