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": 76168.541076954,
-      "required_pages_per_second_upper_estimate": 0.006602096093147939
+      "required_token_positions_per_second": 75460.29326143132,
+      "required_pages_per_second_upper_estimate": 0.00654070696753611
     },
     {
       "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": 9521.06763461925,
-      "required_pages_per_second_upper_estimate": 0.003798096232096398
+      "required_token_positions_per_second": 9432.536657678915,
+      "required_pages_per_second_upper_estimate": 0.0037627799017388363
     },
     {
       "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": 19042.1352692385,
-      "required_pages_per_second_upper_estimate": 0.0044050628644037035
+      "required_token_positions_per_second": 18865.07331535783,
+      "required_pages_per_second_upper_estimate": 0.004364102697557368
     },
     {
       "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": 28563.20290385775,
-      "required_pages_per_second_upper_estimate": 0.0017393491973010562
+      "required_token_positions_per_second": 28297.609973036746,
+      "required_pages_per_second_upper_estimate": 0.001723175981272473
     },
     {
       "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": 28563.20290385775,
-      "required_pages_per_second_upper_estimate": 0.002198874427545185
+      "required_token_positions_per_second": 28297.609973036746,
+      "required_pages_per_second_upper_estimate": 0.002178428348522296
     },
     {
       "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": 28563.20290385775,
-      "required_pages_per_second_upper_estimate": 0.3546593851752331
+      "required_token_positions_per_second": 28297.609973036746,
+      "required_pages_per_second_upper_estimate": 0.3513616098567956
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.1858868573513064,
-    "producer_p95_seconds": 0.38164566599880345,
-    "gpu_update_seconds": 0.9679586737195308,
-    "producer_batches_per_second": 5.379616473423435,
-    "gpu_batches_per_second": 1.033101956881429,
-    "producer_headroom": 5.207246426734688,
+    "producer_batch_seconds": 0.27171062834321447,
+    "producer_p95_seconds": 0.36095523800031515,
+    "gpu_update_seconds": 0.9770436452528775,
+    "producer_batches_per_second": 3.680386027214358,
+    "gpu_batches_per_second": 1.0234957310849517,
+    "producer_headroom": 3.595897779967272,
     "sustainable": true,
-    "buffer_seconds": 60.01343777061091,
+    "buffer_seconds": 60.5767060056784,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391479.75634090806,
-    "upload_bytes_per_second": 5398078.454675434,
+    "generation_bytes_per_second": 376580.0345526598,
+    "upload_bytes_per_second": 5607929.487116227,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.00244252844269,
+    "estimated_full_shards_per_hour": 80.80530908045621,
     "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": 3.5161270839998906,
-      "bytes": 6159420,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.2511098670001957,
       "bytes": 8306730,
@@ -562,6 +556,12 @@
       "bytes": 13453015,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 1.2355715770027018,
+      "bytes": 9750066,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -383,3 +383,4 @@
 {"step": 393300, "loss": 0.5766168162226677, "lr": 0.0003, "grad_norm": 0.16535435980387686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57568269703, "tokens_per_second": 146431.22659633058, "data_wait_s": 1.585899735800922e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 393400, "loss": 0.5997443683445454, "lr": 0.0003, "grad_norm": 0.20393893049244358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57582797421, "tokens_per_second": 149788.42920753785, "data_wait_s": 1.2819997209589928e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 393500, "loss": 0.5762318316847086, "lr": 0.0003, "grad_norm": 0.15132665752762026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57597475623, "tokens_per_second": 149733.49687261274, "data_wait_s": 9.641000360716134e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 393600, "loss": 0.5587992034852505, "lr": 0.0003, "grad_norm": 0.14134306857845638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57612020876, "tokens_per_second": 148340.03096399186, "data_wait_s": 1.0613999620545655e-05, "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:b82a62e735852be083d8e43fb154c2fdd6fb5ea1073c6b096f4a849e86b567b8
-size 821076
+oid sha256:e299a99156fa72c5f0908b64894a66a1dec96691049e69fc9a1d463b97ed0626
+size 823194