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": 75021.72575236797,
-      "required_pages_per_second_upper_estimate": 0.006502693047389688
+      "required_token_positions_per_second": 75142.15341941948,
+      "required_pages_per_second_upper_estimate": 0.006513131412348582
     },
     {
       "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": 9377.715719045997,
-      "required_pages_per_second_upper_estimate": 0.0037409110096720904
+      "required_token_positions_per_second": 9392.769177427435,
+      "required_pages_per_second_upper_estimate": 0.0037469160592897063
     },
     {
       "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": 18755.431438091993,
-      "required_pages_per_second_upper_estimate": 0.004338738978883052
+      "required_token_positions_per_second": 18785.53835485487,
+      "required_pages_per_second_upper_estimate": 0.0043457036842125055
     },
     {
       "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": 28133.147157137988,
-      "required_pages_per_second_upper_estimate": 0.0017131610586539475
+      "required_token_positions_per_second": 28178.307532282302,
+      "required_pages_per_second_upper_estimate": 0.0017159110885620595
     },
     {
       "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": 28133.147157137988,
-      "required_pages_per_second_upper_estimate": 0.0021657675456921975
+      "required_token_positions_per_second": 28178.307532282302,
+      "required_pages_per_second_upper_estimate": 0.0021692441163827262
     },
     {
       "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": 28133.147157137988,
-      "required_pages_per_second_upper_estimate": 0.34931953210497024
+      "required_token_positions_per_second": 28178.307532282302,
+      "required_pages_per_second_upper_estimate": 0.3498802728222097
     }
   ],
   "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": 0.07094002519542641,
-    "producer_p95_seconds": 0.2610363910025626,
-    "gpu_update_seconds": 0.9827553186841061,
-    "producer_batches_per_second": 14.096414502887308,
-    "gpu_batches_per_second": 1.0175472785423174,
-    "producer_headroom": 13.853326327088272,
+    "producer_batch_seconds": 0.11331455633578003,
+    "producer_p95_seconds": 0.29890855999838095,
+    "gpu_update_seconds": 0.9811802915531829,
+    "producer_batches_per_second": 8.824991530979869,
+    "gpu_batches_per_second": 1.0191806833146089,
+    "producer_headroom": 8.658907763321197,
     "sustainable": true,
-    "buffer_seconds": 60.93082975841458,
+    "buffer_seconds": 60.83317807629734,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 160,
+    "upload_pages_per_commit": 158,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 613109.6600796015,
-    "upload_bytes_per_second": 5160535.6124533,
+    "generation_bytes_per_second": 610087.8005970736,
+    "upload_bytes_per_second": 5107056.085674533,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.77953029532925,
+    "estimated_full_shards_per_hour": 65.45531994549826,
     "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": 0.9696599159997277,
-      "bytes": 6181281,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0534396790008032,
       "bytes": 2069289,
@@ -562,6 +556,12 @@
       "bytes": 12128568,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.8681122560010408,
+      "bytes": 1227302,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -231,3 +231,4 @@
 {"step": 63100, "loss": 0.6327682584524155, "lr": 0.0003, "grad_norm": 0.12963534204895175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9234818345, "tokens_per_second": 150186.48587605206, "data_wait_s": 1.1703999916790053e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 63200, "loss": 0.628583986312151, "lr": 0.0003, "grad_norm": 0.1322800532738404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9249359696, "tokens_per_second": 147722.71831464817, "data_wait_s": 9.333001798950136e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 63300, "loss": 0.6152825839817524, "lr": 0.0003, "grad_norm": 0.13509051985736967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9264031394, "tokens_per_second": 149876.53349747547, "data_wait_s": 1.4135999663267285e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 63400, "loss": 0.7246140874922276, "lr": 0.0003, "grad_norm": 0.17346604542955618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9278631187, "tokens_per_second": 148911.84665287365, "data_wait_s": 1.4029999874765053e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:11f978ca00c5ab2a6bea2c4a49fb517caebb184ac3d5d4c11781889ca193197d
-size 496791
+oid sha256:ecb06632a0f25f1eb4159cd4589fc791e8d1563daa3578441221af00d48e754f
+size 498909