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": 75629.75473132246,
-      "required_pages_per_second_upper_estimate": 0.006555395458250106
+      "required_token_positions_per_second": 75628.38869600375,
+      "required_pages_per_second_upper_estimate": 0.006555277053770864
     },
     {
       "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": 9453.719341415308,
-      "required_pages_per_second_upper_estimate": 0.0037712299909906286
+      "required_token_positions_per_second": 9453.548587000469,
+      "required_pages_per_second_upper_estimate": 0.003771161874501543
     },
     {
       "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": 18907.438682830616,
-      "required_pages_per_second_upper_estimate": 0.004373903179717195
+      "required_token_positions_per_second": 18907.097174000937,
+      "required_pages_per_second_upper_estimate": 0.004373824177659799
     },
     {
       "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": 28361.158024245928,
-      "required_pages_per_second_upper_estimate": 0.0017270457241802548
+      "required_token_positions_per_second": 28360.645761001408,
+      "required_pages_per_second_upper_estimate": 0.0017270145300363011
     },
     {
       "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": 28361.158024245928,
-      "required_pages_per_second_upper_estimate": 0.0021833204534166435
+      "required_token_positions_per_second": 28360.645761001408,
+      "required_pages_per_second_upper_estimate": 0.002183281017974045
     },
     {
       "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": 28361.158024245928,
-      "required_pages_per_second_upper_estimate": 0.3521506639711676
+      "required_token_positions_per_second": 28360.645761001408,
+      "required_pages_per_second_upper_estimate": 0.35214430337610547
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12623829424222777,
-    "producer_p95_seconds": 0.270730641001137,
-    "gpu_update_seconds": 0.9748544109645929,
-    "producer_batches_per_second": 7.921526554225981,
-    "gpu_batches_per_second": 1.0257941993723207,
-    "producer_headroom": 7.72233510296035,
+    "producer_batch_seconds": 0.06744973229683637,
+    "producer_p95_seconds": 0.22565372499957448,
+    "gpu_update_seconds": 0.9748720192407831,
+    "producer_batches_per_second": 14.825855729110188,
+    "gpu_batches_per_second": 1.0257756713325161,
+    "producer_headroom": 14.453311911610182,
     "sustainable": true,
-    "buffer_seconds": 60.44097347980476,
+    "buffer_seconds": 60.44206519292855,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329530.13698243233,
-    "upload_bytes_per_second": 5297156.953292832,
+    "generation_bytes_per_second": 342753.1245784973,
+    "upload_bytes_per_second": 5254260.278932557,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.7094963274453,
+    "estimated_full_shards_per_hour": 73.54684165016356,
     "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": 1.2417195279995212,
-      "bytes": 5623378,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 0.9634915199985699,
       "bytes": 7597611,
@@ -562,6 +556,12 @@
       "bytes": 7984887,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0015397960014525,
+      "bytes": 971215,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -162,3 +162,4 @@
 {"step": 211200, "loss": 0.6435531042516232, "lr": 0.0003, "grad_norm": 0.16996013089216458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30912504703, "tokens_per_second": 150602.83732676326, "data_wait_s": 1.7579000996192917e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 211300, "loss": 0.6035649683326483, "lr": 0.0003, "grad_norm": 0.1671182665905871, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30927287802, "tokens_per_second": 150791.66435062894, "data_wait_s": 1.540799712529406e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 211400, "loss": 0.5888801328837872, "lr": 0.0003, "grad_norm": 0.16051327812007865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30941897769, "tokens_per_second": 148663.0822992475, "data_wait_s": 9.882998710963875e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 211500, "loss": 0.6214869990944862, "lr": 0.0003, "grad_norm": 0.1705022461807806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30956467534, "tokens_per_second": 147575.71681998717, "data_wait_s": 1.6814999980852008e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd664309a6267885c27cdceae1835181aefa513f721d1ff934ea6c7e403c729f
-size 349866
+oid sha256:c317d158786592b6624133e43329f7562bb21581cf062883c496f7e945bc6e97
+size 351984