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": 76815.02727785442,
-      "required_pages_per_second_upper_estimate": 0.005314241822501301
+      "required_token_positions_per_second": 76203.33402308331,
+      "required_pages_per_second_upper_estimate": 0.005271923463812341
     },
     {
       "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": 9601.878409731802,
-      "required_pages_per_second_upper_estimate": 0.004020105980132756
+      "required_token_positions_per_second": 9525.416752885414,
+      "required_pages_per_second_upper_estimate": 0.003988093081112135
     },
     {
       "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": 19203.756819463604,
-      "required_pages_per_second_upper_estimate": 0.003999636108370518
+      "required_token_positions_per_second": 19050.83350577083,
+      "required_pages_per_second_upper_estimate": 0.003967786214987293
     },
     {
       "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": 28805.63522919541,
-      "required_pages_per_second_upper_estimate": 0.0017268538483320062
+      "required_token_positions_per_second": 28576.250258656244,
+      "required_pages_per_second_upper_estimate": 0.0017131025695987531
     },
     {
       "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": 28805.63522919541,
-      "required_pages_per_second_upper_estimate": 0.0022395114364709133
+      "required_token_positions_per_second": 28576.250258656244,
+      "required_pages_per_second_upper_estimate": 0.0022216777639693487
     },
     {
       "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": 28805.63522919541,
-      "required_pages_per_second_upper_estimate": 0.36948775963873487
+      "required_token_positions_per_second": 28576.250258656244,
+      "required_pages_per_second_upper_estimate": 0.3665454555310507
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07813074047657409,
-    "producer_p95_seconds": 0.1505852310001501,
-    "gpu_update_seconds": 0.959812195774037,
-    "producer_batches_per_second": 12.799059549420623,
-    "gpu_batches_per_second": 1.0418704871670792,
-    "producer_headroom": 12.284693449972064,
+    "producer_batch_seconds": 0.17847779798437102,
+    "producer_p95_seconds": 0.1601968999998462,
+    "gpu_update_seconds": 0.96751672279413,
+    "producer_batches_per_second": 5.602937795588268,
+    "gpu_batches_per_second": 1.033573866415518,
+    "producer_headroom": 5.420936014006928,
     "sustainable": true,
-    "buffer_seconds": 60.46816833376433,
+    "buffer_seconds": 60.95355353603019,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365507.46087664476,
+    "generation_bytes_per_second": 402710.5647649409,
     "upload_bytes_per_second": 3790757.4461912797,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.42939252590664,
+    "estimated_full_shards_per_hour": 86.41231257640047,
     "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": [

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

@@ -21,3 +21,4 @@
 {"step": 844600, "loss": 0.5306952595710754, "lr": 3e-05, "grad_norm": 0.3029432781467202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123625929123, "tokens_per_second": 150448.66103590364, "data_wait_s": 1.7395000213582534e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 844700, "loss": 0.4973289966583252, "lr": 3e-05, "grad_norm": 0.2948195132684753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123640450885, "tokens_per_second": 150017.66325107688, "data_wait_s": 1.6368000160582596e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 844800, "loss": 0.4664148986339569, "lr": 3e-05, "grad_norm": 0.29752736495565235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123654938761, "tokens_per_second": 150062.9922451329, "data_wait_s": 1.7485999705968425e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 844900, "loss": 0.48894795775413513, "lr": 3e-05, "grad_norm": 0.29625907752383457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123669544844, "tokens_per_second": 151249.47469693146, "data_wait_s": 1.471799987484701e-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 sha256:2a278e301109aa28826765648e3607fc08df4af6fc6d60c10cbe229977a94563
-size 48879
+oid sha256:67747f2108ef7c43e44c664ed955811d8d0ca9e869e3344008154b890012108a
+size 50997