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": 75179.65247888729,
-      "required_pages_per_second_upper_estimate": 0.006516381735782754
+      "required_token_positions_per_second": 76109.09255933682,
+      "required_pages_per_second_upper_estimate": 0.006596943246311771
     },
     {
       "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": 9397.456559860912,
-      "required_pages_per_second_upper_estimate": 0.0037487859262250326
+      "required_token_positions_per_second": 9513.636569917102,
+      "required_pages_per_second_upper_estimate": 0.003795131869282393
     },
     {
       "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": 18794.913119721823,
-      "required_pages_per_second_upper_estimate": 0.004347872371074249
+      "required_token_positions_per_second": 19027.273139834204,
+      "required_pages_per_second_upper_estimate": 0.004401624772330564
     },
     {
       "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": 28192.36967958273,
-      "required_pages_per_second_upper_estimate": 0.0017167673995542706
+      "required_token_positions_per_second": 28540.909709751304,
+      "required_pages_per_second_upper_estimate": 0.0017379916587433237
     },
     {
       "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": 28192.36967958273,
-      "required_pages_per_second_upper_estimate": 0.0021703266594084282
+      "required_token_positions_per_second": 28540.909709751304,
+      "required_pages_per_second_upper_estimate": 0.0021971582357513597
     },
     {
       "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": 28192.36967958273,
-      "required_pages_per_second_upper_estimate": 0.35005487762870147
+      "required_token_positions_per_second": 28540.909709751304,
+      "required_pages_per_second_upper_estimate": 0.3543825783149522
     }
   ],
   "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": 0.10177582812501385,
-    "producer_p95_seconds": 0.19956372800061217,
-    "gpu_update_seconds": 0.9806908860173974,
-    "producer_batches_per_second": 9.825515728269727,
-    "gpu_batches_per_second": 1.0196892968599078,
-    "producer_headroom": 9.635793725134713,
+    "producer_batch_seconds": 0.052114650429693654,
+    "producer_p95_seconds": 0.19034486900000047,
+    "gpu_update_seconds": 0.9687147424930806,
+    "producer_batches_per_second": 19.188462203139416,
+    "gpu_batches_per_second": 1.0322956347566299,
+    "producer_headroom": 18.58814622195241,
     "sustainable": true,
-    "buffer_seconds": 60.802834933078636,
+    "buffer_seconds": 60.060314034571,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272232.89572189085,
-    "upload_bytes_per_second": 4482623.64784836,
+    "generation_bytes_per_second": 272416.2244608926,
+    "upload_bytes_per_second": 4451938.596442435,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.414842164445346,
+    "estimated_full_shards_per_hour": 58.4541802441605,
     "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": [
@@ -490,6 +490,12 @@
       "bytes": 12043048,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 0.7578286200005095,
+      "bytes": 992201,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -57,3 +57,4 @@
 {"step": 5700, "loss": 0.6885922737419605, "lr": 0.0003, "grad_norm": 0.18319819000259852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 833519878, "tokens_per_second": 148250.98255790243, "data_wait_s": 1.0204000318481121e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 5800, "loss": 0.7632534131407738, "lr": 0.0003, "grad_norm": 0.16977647062289472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 848092174, "tokens_per_second": 147800.9207885724, "data_wait_s": 1.1238999832130503e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 5900, "loss": 0.7534791063517332, "lr": 0.0003, "grad_norm": 0.15907890939786187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 862772198, "tokens_per_second": 149378.6033500232, "data_wait_s": 1.3300000318849925e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 6000, "loss": 0.7702555432915688, "lr": 0.0003, "grad_norm": 0.17842406945478048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 877349814, "tokens_per_second": 149471.56779400317, "data_wait_s": 1.1080999684054404e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee4cad2c00bf21d2a43e40726cc89bb3648a6d3e9e28cd8e7482771b0b60a00f
-size 123796
+oid sha256:7dcb8c6d5606e7f3172ea30ec5646cf6358813c0169143b98c05385acd891309
+size 125879