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": 75265.2673453678,
-      "required_pages_per_second_upper_estimate": 0.0065238026154203895
+      "required_token_positions_per_second": 76081.80222637628,
+      "required_pages_per_second_upper_estimate": 0.006594577789417469
     },
     {
       "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": 9408.158418170975,
-      "required_pages_per_second_upper_estimate": 0.0037530550575119574
+      "required_token_positions_per_second": 9510.225278297035,
+      "required_pages_per_second_upper_estimate": 0.0037937710540517935
     },
     {
       "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": 18816.31683634195,
-      "required_pages_per_second_upper_estimate": 0.004352823744221768
+      "required_token_positions_per_second": 19020.45055659407,
+      "required_pages_per_second_upper_estimate": 0.004400046487771323
     },
     {
       "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": 28224.475254512923,
-      "required_pages_per_second_upper_estimate": 0.0017187224606225856
+      "required_token_positions_per_second": 28530.675834891106,
+      "required_pages_per_second_upper_estimate": 0.0017373684694573295
     },
     {
       "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": 28224.475254512923,
-      "required_pages_per_second_upper_estimate": 0.002172798235440474
+      "required_token_positions_per_second": 28530.675834891106,
+      "required_pages_per_second_upper_estimate": 0.002196370403735443
     },
     {
       "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": 28224.475254512923,
-      "required_pages_per_second_upper_estimate": 0.35045352141888725
+      "required_token_positions_per_second": 28530.675834891106,
+      "required_pages_per_second_upper_estimate": 0.3542555078397644
     }
   ],
   "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": 0.11870577537513327,
-    "producer_p95_seconds": 0.35217179200117243,
-    "gpu_update_seconds": 0.9795753419925585,
-    "producer_batches_per_second": 8.42418995065578,
-    "gpu_batches_per_second": 1.0208505228050102,
-    "producer_headroom": 8.252128751923909,
+    "producer_batch_seconds": 0.5012171080389862,
+    "producer_p95_seconds": 0.36003268100103014,
+    "gpu_update_seconds": 0.9690622178037699,
+    "producer_batches_per_second": 1.9951433898824875,
+    "gpu_batches_per_second": 1.0319254859263276,
+    "producer_headroom": 1.933418078236055,
     "sustainable": true,
-    "buffer_seconds": 60.73367120353863,
+    "buffer_seconds": 60.08185750383373,
     "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": 285498.3479631121,
-    "upload_bytes_per_second": 3829894.17726,
+    "generation_bytes_per_second": 264307.0844727497,
+    "upload_bytes_per_second": 3912279.1097512105,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.2612994114878,
+    "estimated_full_shards_per_hour": 56.714147573822665,
     "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": 0.9587836459977552,
-      "bytes": 2593270,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.8898652169991692,
       "bytes": 5769808,
@@ -562,6 +556,12 @@
       "bytes": 4660632,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.4090757320009288,
+      "bytes": 13354980,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -267,3 +267,4 @@
 {"step": 141700, "loss": 0.5687056966125965, "lr": 0.0003, "grad_norm": 0.14378563050977897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20737973026, "tokens_per_second": 150134.11335283951, "data_wait_s": 1.2888998753624037e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 141800, "loss": 0.5475311782211065, "lr": 0.0003, "grad_norm": 0.13591897510819098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20752416941, "tokens_per_second": 146928.80135276602, "data_wait_s": 9.723000403027982e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 141900, "loss": 0.6039933525025845, "lr": 0.0003, "grad_norm": 0.13810004795786504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20766909533, "tokens_per_second": 148248.48222492373, "data_wait_s": 1.1623000318650156e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 142000, "loss": 0.609168142080307, "lr": 0.0003, "grad_norm": 0.14163296518137358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20781541054, "tokens_per_second": 149282.03002226737, "data_wait_s": 1.3641001714859158e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:17d65f615738958bf008f552e4d08e8ee58a568af1a18ed6b3ac32b9df0f4519
-size 573822
+oid sha256:9363582dc11c3385b2d4f1123e54453699265392e72052dfc9b8451a3f098ae9
+size 575940