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": 75820.7524615021,
-      "required_pages_per_second_upper_estimate": 0.006571950657422754
+      "required_token_positions_per_second": 75719.48362746985,
+      "required_pages_per_second_upper_estimate": 0.0065631729315523525
     },
     {
       "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": 9477.594057687762,
-      "required_pages_per_second_upper_estimate": 0.003780753972270529
+      "required_token_positions_per_second": 9464.93545343373,
+      "required_pages_per_second_upper_estimate": 0.0037757042657705965
     },
     {
       "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": 18955.188115375524,
-      "required_pages_per_second_upper_estimate": 0.004384949170575149
+      "required_token_positions_per_second": 18929.87090686746,
+      "required_pages_per_second_upper_estimate": 0.00437909247995446
     },
     {
       "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": 28432.782173063282,
-      "required_pages_per_second_upper_estimate": 0.0017314072590603649
+      "required_token_positions_per_second": 28394.80636030119,
+      "required_pages_per_second_upper_estimate": 0.0017290947313068448
     },
     {
       "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": 28432.782173063282,
-      "required_pages_per_second_upper_estimate": 0.0021888342786609374
+      "required_token_positions_per_second": 28394.80636030119,
+      "required_pages_per_second_upper_estimate": 0.0021859107954707337
     },
     {
       "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": 28432.782173063282,
-      "required_pages_per_second_upper_estimate": 0.3530399961888732
+      "required_token_positions_per_second": 28394.80636030119,
+      "required_pages_per_second_upper_estimate": 0.35256846369123745
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.08019645917184448,
-    "producer_p95_seconds": 0.31875322999985656,
-    "gpu_update_seconds": 0.9723986851414501,
-    "producer_batches_per_second": 12.469378452946485,
-    "gpu_batches_per_second": 1.0283847718845227,
-    "producer_headroom": 12.12520721217629,
+    "producer_batch_seconds": 0.13170573854694112,
+    "producer_p95_seconds": 0.34831243899679976,
+    "gpu_update_seconds": 0.9736991916470576,
+    "producer_batches_per_second": 7.5926835917145015,
+    "gpu_batches_per_second": 1.0270112254159864,
+    "producer_headroom": 7.392989875684289,
     "sustainable": true,
-    "buffer_seconds": 60.28871847876991,
+    "buffer_seconds": 60.369349882117575,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396547.9842474775,
+    "generation_bytes_per_second": 396049.5212092127,
     "upload_bytes_per_second": 3919643.0195153127,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.08996625488514,
+    "estimated_full_shards_per_hour": 84.98300769049918,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 138400, "loss": 0.6374600455164909, "lr": 0.0003, "grad_norm": 0.16114978434019975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20254527171, "tokens_per_second": 148251.54884230747, "data_wait_s": 1.8317001376999542e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 138500, "loss": 0.5999759268015623, "lr": 0.0003, "grad_norm": 0.14957302639950082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20269059703, "tokens_per_second": 147830.30864382494, "data_wait_s": 1.2942000466864556e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 138600, "loss": 0.5904894862323999, "lr": 0.0003, "grad_norm": 0.1377562655743542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20283675767, "tokens_per_second": 149458.7950497035, "data_wait_s": 1.112900281441398e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 138700, "loss": 0.5369933377951384, "lr": 0.0003, "grad_norm": 0.13935597412628523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20298343512, "tokens_per_second": 148685.80107937855, "data_wait_s": 9.620001947041601e-06, "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 sha256:1b430e754ca13c75037689f409e12b618b88fa99ad689565b97f9e6e6de3c6c1
-size 503145
+oid sha256:f3902f7923c1cb2ccb628d8db83a60be0485e3872387a9f8faab17b00719441e
+size 505263