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": 76236.29869072248,
-      "required_pages_per_second_upper_estimate": 0.0066079691513792884
+      "required_token_positions_per_second": 76306.56527147602,
+      "required_pages_per_second_upper_estimate": 0.0066140596805099846
     },
     {
       "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": 9529.53733634031,
-      "required_pages_per_second_upper_estimate": 0.0038014749227462544
+      "required_token_positions_per_second": 9538.320658934503,
+      "required_pages_per_second_upper_estimate": 0.003804978721451453
     },
     {
       "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": 19059.07467268062,
-      "required_pages_per_second_upper_estimate": 0.00440898149726672
+      "required_token_positions_per_second": 19076.641317869005,
+      "required_pages_per_second_upper_estimate": 0.004413045231468657
     },
     {
       "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": 28588.612009020933,
-      "required_pages_per_second_upper_estimate": 0.0017408964784942221
+      "required_token_positions_per_second": 28614.961976803505,
+      "required_pages_per_second_upper_estimate": 0.001742501053284584
     },
     {
       "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": 28588.612009020933,
-      "required_pages_per_second_upper_estimate": 0.002200830490797552
+      "required_token_positions_per_second": 28614.961976803505,
+      "required_pages_per_second_upper_estimate": 0.0022028589842588335
     },
     {
       "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": 28588.612009020933,
-      "required_pages_per_second_upper_estimate": 0.35497488122255527
+      "required_token_positions_per_second": 28614.961976803505,
+      "required_pages_per_second_upper_estimate": 0.3553020596347456
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09750765496875147,
-    "producer_p95_seconds": 0.2419827890007582,
-    "gpu_update_seconds": 0.967098367394537,
-    "producer_batches_per_second": 10.255605063217576,
-    "gpu_batches_per_second": 1.0340209783355372,
-    "producer_headroom": 9.918178913280865,
+    "producer_batch_seconds": 0.1089492368281526,
+    "producer_p95_seconds": 0.19254466900019906,
+    "gpu_update_seconds": 0.9662078189169929,
+    "producer_batches_per_second": 9.178586551985822,
+    "gpu_batches_per_second": 1.0349740298323027,
+    "producer_headroom": 8.868422093135063,
     "sustainable": true,
-    "buffer_seconds": 60.927197145855835,
+    "buffer_seconds": 60.871092591770555,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317347.3708830839,
-    "upload_bytes_per_second": 5584406.092897304,
+    "generation_bytes_per_second": 347446.2962334346,
+    "upload_bytes_per_second": 5599964.374492737,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.09535832280528,
+    "estimated_full_shards_per_hour": 74.55388703586844,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 1099454,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.1156518739990133,
+      "bytes": 7370564,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -74,3 +74,4 @@
 {"step": 287600, "loss": 0.5928755290806293, "lr": 0.0003, "grad_norm": 0.15635010854087456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42096763738, "tokens_per_second": 125391.40559606963, "data_wait_s": 8.151000656653196e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 287700, "loss": 0.5897219739854336, "lr": 0.0003, "grad_norm": 0.15187574616742733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42111467830, "tokens_per_second": 150928.29858113185, "data_wait_s": 1.4171000657370314e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 287800, "loss": 0.5792989619076252, "lr": 0.0003, "grad_norm": 0.15761427272184975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42125903408, "tokens_per_second": 148096.6619555482, "data_wait_s": 9.186000170302577e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 287900, "loss": 0.5967279635369778, "lr": 0.0003, "grad_norm": 0.1422105488783368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42140400954, "tokens_per_second": 148699.04759279452, "data_wait_s": 1.422199966327753e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6d1fc8a3aa3b56f9f503d03998c03578d4982e4d8c5bb2adbfec71b20b84c60
-size 161916
+oid sha256:d17ffee25ab6dcc618f3d1bdb9fc5ae0702b3a156386e7594de5202da6535946
+size 164034