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": 75978.40953900713,
-      "required_pages_per_second_upper_estimate": 0.006585615973322681
+      "required_token_positions_per_second": 76047.83361425264,
+      "required_pages_per_second_upper_estimate": 0.006591633476211251
     },
     {
       "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": 9497.301192375891,
-      "required_pages_per_second_upper_estimate": 0.0037886154429455448
+      "required_token_positions_per_second": 9505.97920178158,
+      "required_pages_per_second_upper_estimate": 0.003792077230645277
     },
     {
       "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": 18994.602384751783,
-      "required_pages_per_second_upper_estimate": 0.004394066968127897
+      "required_token_positions_per_second": 19011.95840356316,
+      "required_pages_per_second_upper_estimate": 0.00439808197762441
     },
     {
       "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": 28491.903577127676,
-      "required_pages_per_second_upper_estimate": 0.0017350074423818519
+      "required_token_positions_per_second": 28517.937605344738,
+      "required_pages_per_second_upper_estimate": 0.001736592778110281
     },
     {
       "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": 28491.903577127676,
-      "required_pages_per_second_upper_estimate": 0.002193385608004335
+      "required_token_positions_per_second": 28517.937605344738,
+      "required_pages_per_second_upper_estimate": 0.0021953897795632875
     },
     {
       "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": 28491.903577127676,
-      "required_pages_per_second_upper_estimate": 0.35377408616074196
+      "required_token_positions_per_second": 28517.937605344738,
+      "required_pages_per_second_upper_estimate": 0.35409734166091034
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.0863802220235641,
-    "producer_p95_seconds": 0.31177999899955466,
-    "gpu_update_seconds": 0.9703809338381613,
-    "producer_batches_per_second": 11.576724122417803,
-    "gpu_batches_per_second": 1.030523132853287,
-    "producer_headroom": 11.233832364698554,
+    "producer_batch_seconds": 0.08794142853926701,
+    "producer_p95_seconds": 0.28989767500024755,
+    "gpu_update_seconds": 0.969495072982357,
+    "producer_batches_per_second": 11.371204864536477,
+    "gpu_batches_per_second": 1.0314647571377582,
+    "producer_headroom": 11.024327090041126,
     "sustainable": true,
-    "buffer_seconds": 60.163617897966,
+    "buffer_seconds": 60.108694524906134,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317872.3206697834,
-    "upload_bytes_per_second": 3969449.7594727287,
+    "generation_bytes_per_second": 330847.88517483306,
+    "upload_bytes_per_second": 4013468.448973137,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.2080003268254,
+    "estimated_full_shards_per_hour": 70.99225441392655,
     "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": 1.1932972649992735,
-      "bytes": 7607279,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.4406277779999073,
       "bytes": 9581963,
@@ -562,6 +556,12 @@
       "bytes": 669079,
       "pages": 20,
       "shards": 1
+    },
+    {
+      "seconds": 1.4896883629990043,
+      "bytes": 13505965,
+      "pages": 48,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -140,3 +140,4 @@
 {"step": 294200, "loss": 0.633145309984684, "lr": 0.0003, "grad_norm": 0.153817176875024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43062718440, "tokens_per_second": 149666.37332655812, "data_wait_s": 8.15800012787804e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 294300, "loss": 0.5643345396965742, "lr": 0.0003, "grad_norm": 0.148157780776155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43077410234, "tokens_per_second": 139069.29664660958, "data_wait_s": 1.1342999641783535e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 294400, "loss": 0.6025479212403297, "lr": 0.0003, "grad_norm": 0.1530220006729937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43092112910, "tokens_per_second": 150859.8605852468, "data_wait_s": 9.805000445339829e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 294500, "loss": 0.6064490936696529, "lr": 0.0003, "grad_norm": 0.15015753211878693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43106871866, "tokens_per_second": 151812.81842149483, "data_wait_s": 1.2907999916933477e-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 sha256:8596ebd46aabdafbd4d6800298517c84e8adf136869a32b8c139bad06bef6556
-size 302487
+oid sha256:db1cc2d329aef6abf906f5474fed3fa5412f8a4ae0258489d19af1b7c1d3b144
+size 304605