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": 76418.88780981897,
-      "required_pages_per_second_upper_estimate": 0.006623795513454681
+      "required_token_positions_per_second": 75495.09424515428,
+      "required_pages_per_second_upper_estimate": 0.006543723428602929
     },
     {
       "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": 9552.360976227372,
-      "required_pages_per_second_upper_estimate": 0.0038105796139410292
+      "required_token_positions_per_second": 9436.886780644285,
+      "required_pages_per_second_upper_estimate": 0.003764515230829857
     },
     {
       "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": 19104.721952454744,
-      "required_pages_per_second_upper_estimate": 0.004419541192077777
+      "required_token_positions_per_second": 18873.77356128857,
+      "required_pages_per_second_upper_estimate": 0.004366115346334346
     },
     {
       "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": 28657.082928682117,
-      "required_pages_per_second_upper_estimate": 0.0017450659982624363
+      "required_token_positions_per_second": 28310.660341932857,
+      "required_pages_per_second_upper_estimate": 0.0017239706802682521
     },
     {
       "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": 28657.082928682117,
-      "required_pages_per_second_upper_estimate": 0.0022061015717327066
+      "required_token_positions_per_second": 28310.660341932857,
+      "required_pages_per_second_upper_estimate": 0.002179433002045652
     },
     {
       "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": 28657.082928682117,
-      "required_pages_per_second_upper_estimate": 0.35582506088732035
+      "required_token_positions_per_second": 28310.660341932857,
+      "required_pages_per_second_upper_estimate": 0.35152365176171024
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.05022175149999519,
-    "producer_p95_seconds": 0.1299934950000079,
-    "gpu_update_seconds": 0.9647876606564113,
-    "producer_batches_per_second": 19.911691052831873,
-    "gpu_batches_per_second": 1.0364975017607827,
-    "producer_headroom": 19.21055383057486,
+    "producer_batch_seconds": 0.045304454437493646,
+    "producer_p95_seconds": 0.12189184299995759,
+    "gpu_update_seconds": 0.9765932573126405,
+    "producer_batches_per_second": 22.072884717764243,
+    "gpu_batches_per_second": 1.0239677496358817,
+    "producer_headroom": 21.556230384807787,
     "sustainable": true,
-    "buffer_seconds": 60.78162262135391,
+    "buffer_seconds": 60.54878195338371,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 58,
+    "upload_pages_per_commit": 63,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -8,3 +8,4 @@
 {"step": 355800, "loss": 0.6531054731458426, "lr": 0.0003, "grad_norm": 0.17719566197741718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52079015267, "tokens_per_second": 148980.5927603168, "data_wait_s": 1.1954000001423992e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 355900, "loss": 0.5476674549281597, "lr": 0.0003, "grad_norm": 0.1451582331959875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52093511172, "tokens_per_second": 149512.22858483213, "data_wait_s": 1.1758000027839444e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356000, "loss": 0.5913226958364248, "lr": 0.0003, "grad_norm": 0.16621256028653167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52108078318, "tokens_per_second": 149870.95399072525, "data_wait_s": 1.1811000149464235e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 356100, "loss": 0.5715948585420847, "lr": 0.0003, "grad_norm": 0.14480262907768615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52122825287, "tokens_per_second": 151006.9257437939, "data_wait_s": 7.612999979755841e-06, "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 sha256:8954edc84b2cde0d0969acd08c27ce4276f77c21376f3c6642717e9f944a06ab
-size 21345
+oid sha256:25cf40b99b614710e0b2d049994452b756810f3ebecb2c869c9d4c84fd509053
+size 23463