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": 75769.68501589732,
-      "required_pages_per_second_upper_estimate": 0.0065675242606669905
+      "required_token_positions_per_second": 75866.03401292826,
+      "required_pages_per_second_upper_estimate": 0.006575875547535331
     },
     {
       "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": 9471.210626987166,
-      "required_pages_per_second_upper_estimate": 0.003778207526323267
+      "required_token_positions_per_second": 9483.254251616032,
+      "required_pages_per_second_upper_estimate": 0.0037830119082559567
     },
     {
       "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": 18942.42125397433,
-      "required_pages_per_second_upper_estimate": 0.00438199578187907
+      "required_token_positions_per_second": 18966.508503232064,
+      "required_pages_per_second_upper_estimate": 0.004387567943073737
     },
     {
       "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": 28413.6318809615,
-      "required_pages_per_second_upper_estimate": 0.001730241106745184
+      "required_token_positions_per_second": 28449.762754848096,
+      "required_pages_per_second_upper_estimate": 0.0017324412874008327
     },
     {
       "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": 28413.6318809615,
-      "required_pages_per_second_upper_estimate": 0.0021873600361635407
+      "required_token_positions_per_second": 28449.762754848096,
+      "required_pages_per_second_upper_estimate": 0.002190141490851991
     },
     {
       "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": 28413.6318809615,
-      "required_pages_per_second_upper_estimate": 0.35280221365287384
+      "required_token_positions_per_second": 28449.762754848096,
+      "required_pages_per_second_upper_estimate": 0.35325083818428915
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.126220250156166,
-    "producer_p95_seconds": 0.31565818400122225,
-    "gpu_update_seconds": 0.9730540648879699,
-    "producer_batches_per_second": 7.922658993012215,
-    "gpu_batches_per_second": 1.0276921253241282,
-    "producer_headroom": 7.709175537871766,
+    "producer_batch_seconds": 0.10830060166395583,
+    "producer_p95_seconds": 0.3221580220015312,
+    "gpu_update_seconds": 0.9718182973349587,
+    "producer_batches_per_second": 9.23355904432446,
+    "gpu_batches_per_second": 1.0289989422326422,
+    "producer_headroom": 8.973341628797206,
     "sustainable": true,
-    "buffer_seconds": 60.329352023054135,
+    "buffer_seconds": 60.25273443476744,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 159,
+    "upload_pages_per_commit": 171,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 583766.7083149065,
+    "generation_bytes_per_second": 570720.7222349446,
     "upload_bytes_per_second": 2627814.0201740037,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.631373105456326,
+    "estimated_full_shards_per_hour": 61.231690646582855,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -204,3 +204,4 @@
 {"step": 300600, "loss": 0.6378210633993149, "lr": 0.0003, "grad_norm": 0.14847135746426773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44000313686, "tokens_per_second": 150242.5736800298, "data_wait_s": 1.339799928246066e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 300700, "loss": 0.5964351166039705, "lr": 0.0003, "grad_norm": 0.1417063265200555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44014786981, "tokens_per_second": 147945.62333285174, "data_wait_s": 1.1845997505588457e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 300800, "loss": 0.6383797377347946, "lr": 0.0003, "grad_norm": 0.19675669623382952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44029298312, "tokens_per_second": 149009.68773356857, "data_wait_s": 9.657996997702867e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 300900, "loss": 0.6143244709819555, "lr": 0.0003, "grad_norm": 0.15830787618421865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44043880447, "tokens_per_second": 149355.67107736587, "data_wait_s": 1.402800262439996e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:105efbcd45e47eb97afcb220cc652a7ba6b48ddb6ea87796b8775b0214f73f73
-size 439605
+oid sha256:256769ca64e50b359c16237249369c42bb447c24cea6053184421cc94b0b02dc
+size 441723