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": 77225.93846131285,
-      "required_pages_per_second_upper_estimate": 0.005342669611618261
+      "required_token_positions_per_second": 77096.75342299492,
+      "required_pages_per_second_upper_estimate": 0.00533373229091684
     },
     {
       "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": 9653.242307664106,
-      "required_pages_per_second_upper_estimate": 0.0040416109715968535
+      "required_token_positions_per_second": 9637.094177874365,
+      "required_pages_per_second_upper_estimate": 0.004034850086865184
     },
     {
       "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": 19306.484615328212,
-      "required_pages_per_second_upper_estimate": 0.004021031599134959
+      "required_token_positions_per_second": 19274.18835574873,
+      "required_pages_per_second_upper_estimate": 0.004014305139974193
     },
     {
       "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": 28959.72692299232,
-      "required_pages_per_second_upper_estimate": 0.001736091410090738
+      "required_token_positions_per_second": 28911.282533623093,
+      "required_pages_per_second_upper_estimate": 0.0017331872429183523
     },
     {
       "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": 28959.72692299232,
-      "required_pages_per_second_upper_estimate": 0.002251491387885892
+      "required_token_positions_per_second": 28911.282533623093,
+      "required_pages_per_second_upper_estimate": 0.002247725049696784
     },
     {
       "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": 28959.72692299232,
-      "required_pages_per_second_upper_estimate": 0.37146428243599133
+      "required_token_positions_per_second": 28911.282533623093,
+      "required_pages_per_second_upper_estimate": 0.3708428898246956
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.3068276530545404,
-    "producer_p95_seconds": 0.30582650799988187,
-    "gpu_update_seconds": 0.9547051349454927,
-    "producer_batches_per_second": 3.2591586515907816,
-    "gpu_batches_per_second": 1.0474438267864699,
-    "producer_headroom": 3.111535500275747,
+    "producer_batch_seconds": 0.09957253992183723,
+    "producer_p95_seconds": 0.30079320200093207,
+    "gpu_update_seconds": 0.9563048601474553,
+    "producer_batches_per_second": 10.04292951435188,
+    "gpu_batches_per_second": 1.0456916425644927,
+    "producer_headroom": 9.604102304693026,
     "sustainable": true,
-    "buffer_seconds": 60.14642350156604,
+    "buffer_seconds": 60.247206189289685,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 376490.1481785479,
+    "generation_bytes_per_second": 362601.6589852402,
     "upload_bytes_per_second": 5789703.918124365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.7860215570195,
+    "estimated_full_shards_per_hour": 77.80587508361725,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -290,3 +290,4 @@
 {"step": 586100, "loss": 0.6081873774528503, "lr": 0.0003, "grad_norm": 0.18081899728163095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85787713849, "tokens_per_second": 153397.38388069393, "data_wait_s": 1.221899947267957e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 586200, "loss": 0.6391074657440186, "lr": 0.0003, "grad_norm": 0.18716523192384205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85802520527, "tokens_per_second": 153722.66691978186, "data_wait_s": 1.5544999769190326e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 586300, "loss": 0.5808490514755249, "lr": 0.0003, "grad_norm": 0.17973211195187186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85817229576, "tokens_per_second": 152600.74465313466, "data_wait_s": 1.2349999451544136e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 586400, "loss": 0.5779256820678711, "lr": 0.0003, "grad_norm": 0.16024750350676212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85831748706, "tokens_per_second": 150906.63926133554, "data_wait_s": 1.4175999240251258e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd127151669398d70fbf13b89de49c0b91a4dc2c9f221ef792d0a4a34a7cdf28
-size 623319
+oid sha256:d0aea38e108417d78309f8834e012bd6a6b7c0963c8d205c4f77fc15f24a3986
+size 625437