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": 76828.87211048772,
-      "required_pages_per_second_upper_estimate": 0.0053151996401473115
+      "required_token_positions_per_second": 76214.46440820044,
+      "required_pages_per_second_upper_estimate": 0.005272693489680795
     },
     {
       "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": 9603.609013810965,
-      "required_pages_per_second_upper_estimate": 0.004020830547921578
+      "required_token_positions_per_second": 9526.808051025055,
+      "required_pages_per_second_upper_estimate": 0.003988675588589593
     },
     {
       "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": 19207.21802762193,
-      "required_pages_per_second_upper_estimate": 0.0040003569867516995
+      "required_token_positions_per_second": 19053.61610205011,
+      "required_pages_per_second_upper_estimate": 0.003968365756410268
     },
     {
       "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": 28810.827041432894,
-      "required_pages_per_second_upper_estimate": 0.0017271650895482051
+      "required_token_positions_per_second": 28580.42415307516,
+      "required_pages_per_second_upper_estimate": 0.0017133527881959994
     },
     {
       "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": 28810.827041432894,
-      "required_pages_per_second_upper_estimate": 0.002239915077036009
+      "required_token_positions_per_second": 28580.42415307516,
+      "required_pages_per_second_upper_estimate": 0.002222002265901398
     },
     {
       "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": 28810.827041432894,
-      "required_pages_per_second_upper_estimate": 0.36955435463158365
+      "required_token_positions_per_second": 28580.42415307516,
+      "required_pages_per_second_upper_estimate": 0.3665989937670779
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.2680531452655259,
-    "producer_p95_seconds": 0.25822584300112794,
-    "gpu_update_seconds": 0.9596392342447986,
-    "producer_batches_per_second": 3.730603492861194,
-    "gpu_batches_per_second": 1.0420582697277523,
-    "producer_headroom": 3.580033479160287,
+    "producer_batch_seconds": 0.42453274039826283,
+    "producer_p95_seconds": 0.24503954600004363,
+    "gpu_update_seconds": 0.9673754263379315,
+    "producer_batches_per_second": 2.355530928101987,
+    "gpu_batches_per_second": 1.0337248319254615,
+    "producer_headroom": 2.278682735824843,
     "sustainable": true,
-    "buffer_seconds": 60.45727175742231,
+    "buffer_seconds": 60.944651859289685,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438973.81723265315,
-    "upload_bytes_per_second": 5347094.72695548,
+    "generation_bytes_per_second": 395289.408148943,
+    "upload_bytes_per_second": 5458798.309446867,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.19356239065834,
+    "estimated_full_shards_per_hour": 84.81990512229173,
     "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": 0.900488690000202,
-      "bytes": 1094466,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.2211129879997316,
       "bytes": 12049529,
@@ -562,6 +556,12 @@
       "bytes": 6064387,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.2800072050013114,
+      "bytes": 12000660,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -221,3 +221,4 @@
 {"step": 579200, "loss": 0.5281848311424255, "lr": 0.0003, "grad_norm": 0.16683918222658417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84777790977, "tokens_per_second": 150649.3625108852, "data_wait_s": 1.4092001947574317e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 579300, "loss": 0.5518209934234619, "lr": 0.0003, "grad_norm": 0.2045410055942124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84792385921, "tokens_per_second": 152927.73143455957, "data_wait_s": 1.0921001376118511e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 579400, "loss": 0.5096331238746643, "lr": 0.0003, "grad_norm": 0.1611796267319173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84806918627, "tokens_per_second": 150879.75185140574, "data_wait_s": 1.630000042496249e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 579500, "loss": 0.6025071740150452, "lr": 0.0003, "grad_norm": 0.16887081732719386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84821567494, "tokens_per_second": 145632.65346693678, "data_wait_s": 1.22540004667826e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e5aa11c2cb2bf516d2c0955e3d8f407800fca1055a7cf1694bc2d1d3855e298
-size 475611
+oid sha256:38823d471cd7c671cb344a94191cb38440c23158d29205748ed2f6056b15d641
+size 477729