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": 77306.31768074245,
-      "required_pages_per_second_upper_estimate": 0.005348230432523885
+      "required_token_positions_per_second": 76853.46941965164,
+      "required_pages_per_second_upper_estimate": 0.005316901339068892
     },
     {
       "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": 9663.289710092806,
-      "required_pages_per_second_upper_estimate": 0.004045817609180128
+      "required_token_positions_per_second": 9606.683677456454,
+      "required_pages_per_second_upper_estimate": 0.004022117845383667
     },
     {
       "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": 19326.57942018561,
-      "required_pages_per_second_upper_estimate": 0.004025216817051114
+      "required_token_positions_per_second": 19213.36735491291,
+      "required_pages_per_second_upper_estimate": 0.004001637729457745
     },
     {
       "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": 28989.869130278417,
-      "required_pages_per_second_upper_estimate": 0.001737898389393054
+      "required_token_positions_per_second": 28820.051032369363,
+      "required_pages_per_second_upper_estimate": 0.0017277180537206278
     },
     {
       "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": 28989.869130278417,
-      "required_pages_per_second_upper_estimate": 0.0022538348119208276
+      "required_token_positions_per_second": 28820.051032369363,
+      "required_pages_per_second_upper_estimate": 0.0022406322017592714
     },
     {
       "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": 28989.869130278417,
-      "required_pages_per_second_upper_estimate": 0.3718509143068767
+      "required_token_positions_per_second": 28820.051032369363,
+      "required_pages_per_second_upper_estimate": 0.36967267008336685
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11347129023454272,
-    "producer_p95_seconds": 0.33404291300394107,
-    "gpu_update_seconds": 0.9537124805825045,
-    "producer_batches_per_second": 8.812801880837183,
-    "gpu_batches_per_second": 1.0485340397236118,
-    "producer_headroom": 8.40487914265539,
+    "producer_batch_seconds": 0.1538562223830695,
+    "producer_p95_seconds": 0.32372215500072343,
+    "gpu_update_seconds": 0.9593320972591974,
+    "producer_batches_per_second": 6.49957463215372,
+    "gpu_batches_per_second": 1.0423918920851187,
+    "producer_headroom": 6.235250563156705,
     "sustainable": true,
-    "buffer_seconds": 60.08388627669778,
+    "buffer_seconds": 60.437922127329436,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266109.91867260274,
-    "upload_bytes_per_second": 4060262.8605182804,
+    "generation_bytes_per_second": 294939.1491264792,
+    "upload_bytes_per_second": 4061949.122925757,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.10099382527887,
+    "estimated_full_shards_per_hour": 63.287075570543124,
     "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": 1.058052360000147,
-      "bytes": 885421,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.6197210989994346,
       "bytes": 11971576,
@@ -562,6 +556,12 @@
       "bytes": 6146671,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.0191766480056685,
+      "bytes": 906488,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -351,3 +351,4 @@
 {"step": 877600, "loss": 0.5132738947868347, "lr": 3e-05, "grad_norm": 0.34010128533589473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128456356294, "tokens_per_second": 151670.18337014448, "data_wait_s": 1.408400567015633e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 877700, "loss": 0.5507510900497437, "lr": 3e-05, "grad_norm": 0.3248438304098416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128471066757, "tokens_per_second": 153429.1232770866, "data_wait_s": 9.777002560440451e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 877800, "loss": 0.5265238285064697, "lr": 3e-05, "grad_norm": 0.30522589961350977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128485751876, "tokens_per_second": 151791.50730786676, "data_wait_s": 1.0826006473507732e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 877900, "loss": 0.5299591422080994, "lr": 3e-05, "grad_norm": 0.31637334975366893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128500210678, "tokens_per_second": 150146.79126671597, "data_wait_s": 1.4199998986441642e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b35545c103c462e5b3397e23aa18b644dc79a8809764423d47e2b9e2746c2f77
-size 753300
+oid sha256:587549bca7680941369f78a3956eb87acc01180bf86da75bad4b0162dc7d6338
+size 755418