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": 76950.97886476592,
-      "required_pages_per_second_upper_estimate": 0.005323647268734981
+      "required_token_positions_per_second": 76725.10323720779,
+      "required_pages_per_second_upper_estimate": 0.005308020668716847
     },
     {
       "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": 9618.87235809574,
-      "required_pages_per_second_upper_estimate": 0.00402722099143874
+      "required_token_positions_per_second": 9590.637904650974,
+      "required_pages_per_second_upper_estimate": 0.004015399815383851
     },
     {
       "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": 19237.74471619148,
-      "required_pages_per_second_upper_estimate": 0.004006714890937822
+      "required_token_positions_per_second": 19181.275809301947,
+      "required_pages_per_second_upper_estimate": 0.003994953906817951
     },
     {
       "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": 28856.61707428722,
-      "required_pages_per_second_upper_estimate": 0.0017299101321004901
+      "required_token_positions_per_second": 28771.913713952923,
+      "required_pages_per_second_upper_estimate": 0.0017248322949829915
     },
     {
       "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": 28856.61707428722,
-      "required_pages_per_second_upper_estimate": 0.002243475050681364
+      "required_token_positions_per_second": 28771.913713952923,
+      "required_pages_per_second_upper_estimate": 0.00223688973698608
     },
     {
       "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": 28856.61707428722,
-      "required_pages_per_second_upper_estimate": 0.3701417000075322
+      "required_token_positions_per_second": 28771.913713952923,
+      "required_pages_per_second_upper_estimate": 0.36905521624854637
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0950580643906278,
+    "producer_batch_seconds": 0.07292904376561182,
     "producer_p95_seconds": 0.2651801179999893,
-    "gpu_update_seconds": 0.9581164669726944,
-    "producer_batches_per_second": 10.519885991899015,
-    "gpu_batches_per_second": 1.0437144485780967,
-    "producer_headroom": 10.079275999513822,
+    "gpu_update_seconds": 0.9609371234347932,
+    "producer_batches_per_second": 13.71195820438728,
+    "gpu_batches_per_second": 1.0406508143067463,
+    "producer_headroom": 13.176329673582025,
     "sustainable": true,
-    "buffer_seconds": 60.36133741927974,
+    "buffer_seconds": 60.53903877639197,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432870.30047593,
-    "upload_bytes_per_second": 5624009.240021398,
+    "generation_bytes_per_second": 417044.9394404515,
+    "upload_bytes_per_second": 5610870.517087748,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.88388977726387,
+    "estimated_full_shards_per_hour": 89.48813569460067,
     "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": 1.9063561590010067,
-      "bytes": 11430108,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.1905037959986657,
       "bytes": 6976659,
@@ -562,6 +556,12 @@
       "bytes": 6389566,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.1336718419988756,
+      "bytes": 6075801,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -226,3 +226,4 @@
 {"step": 507600, "loss": 0.6018542647361755, "lr": 0.0003, "grad_norm": 0.18386225132672107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74295830453, "tokens_per_second": 151760.04939684746, "data_wait_s": 2.4901000870158896e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 507700, "loss": 0.6152762770652771, "lr": 0.0003, "grad_norm": 0.1614442365486507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74310310597, "tokens_per_second": 150534.92371814157, "data_wait_s": 9.554001735523343e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 507800, "loss": 0.5956807732582092, "lr": 0.0003, "grad_norm": 0.15711850903371788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74324960341, "tokens_per_second": 152888.8229281551, "data_wait_s": 1.4253000699682161e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 507900, "loss": 0.5805848240852356, "lr": 0.0003, "grad_norm": 0.19505596456534124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74339433629, "tokens_per_second": 150216.0210410353, "data_wait_s": 1.0264000593451783e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8daaa3083cb798afafcf1cfbc34b8202d36461bfd897a33773e7c1617e8942bb
-size 486201
+oid sha256:a72349444b0a7299a2e5f739174c11d5118cfd4d9695005718efbbeb286fac76
+size 488319