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": 75123.47751672944,
-      "required_pages_per_second_upper_estimate": 0.006511512632437051
+      "required_token_positions_per_second": 75048.63001254914,
+      "required_pages_per_second_upper_estimate": 0.006505025040473968
     },
     {
       "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": 9390.43468959118,
-      "required_pages_per_second_upper_estimate": 0.0037459847971881206
+      "required_token_positions_per_second": 9381.078751568642,
+      "required_pages_per_second_upper_estimate": 0.003742252573627191
     },
     {
       "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": 18780.86937918236,
-      "required_pages_per_second_upper_estimate": 0.004344623598861836
+      "required_token_positions_per_second": 18762.157503137285,
+      "required_pages_per_second_upper_estimate": 0.004340294935656578
     },
     {
       "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": 28171.30406877354,
-      "required_pages_per_second_upper_estimate": 0.0017154846143787091
+      "required_token_positions_per_second": 28143.23625470593,
+      "required_pages_per_second_upper_estimate": 0.0017137754317624312
     },
     {
       "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": 28171.30406877354,
-      "required_pages_per_second_upper_estimate": 0.002168704970374994
+      "required_token_positions_per_second": 28143.23625470593,
+      "required_pages_per_second_upper_estimate": 0.002166544232351387
     },
     {
       "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": 28171.30406877354,
-      "required_pages_per_second_upper_estimate": 0.34979331324451546
+      "required_token_positions_per_second": 28143.23625470593,
+      "required_pages_per_second_upper_estimate": 0.34944480493072666
     }
   ],
   "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": 0.05738109259381474,
-    "producer_p95_seconds": 0.19426240199936728,
-    "gpu_update_seconds": 0.9814242156665514,
-    "producer_batches_per_second": 17.427343307641944,
-    "gpu_batches_per_second": 1.0189273751726542,
-    "producer_headroom": 17.103616736854217,
+    "producer_batch_seconds": 0.07856175708587898,
+    "producer_p95_seconds": 0.1934798509992106,
+    "gpu_update_seconds": 0.9824030097241173,
+    "producer_batches_per_second": 12.72883954093415,
+    "gpu_batches_per_second": 1.017912190925417,
+    "producer_headroom": 12.50485027530906,
     "sustainable": true,
-    "buffer_seconds": 60.84830137132619,
+    "buffer_seconds": 60.908986602895276,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321926.6827531263,
+    "generation_bytes_per_second": 338496.2381600554,
     "upload_bytes_per_second": 5798834.912783201,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.07797204919187,
+    "estimated_full_shards_per_hour": 72.63341292001006,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 92500, "loss": 0.6128994338214397, "lr": 0.0003, "grad_norm": 0.13664383668330488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13536942201, "tokens_per_second": 148515.6514204053, "data_wait_s": 1.4106999515206553e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 92600, "loss": 0.5336556099355221, "lr": 0.0003, "grad_norm": 0.1350423349771974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13551411504, "tokens_per_second": 147043.00553476688, "data_wait_s": 1.8695998733164743e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 92700, "loss": 0.5997125413268805, "lr": 0.0003, "grad_norm": 0.1257633616212327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13565981359, "tokens_per_second": 147937.23107947153, "data_wait_s": 1.3378999938140623e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 92800, "loss": 0.6019905339926481, "lr": 0.0003, "grad_norm": 0.12667470160401403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13580623119, "tokens_per_second": 149293.2970835035, "data_wait_s": 1.1711999832186848e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d55169e6c4caf4b7204cc312d663bd36005d91132c373b4bfaca2ebc3a6b8c44
-size 164034
+oid sha256:84427ba1cff545dee8b56058f5e6628823435934c12c8a5c5a0b6d3edb2b81c1
+size 166152