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": 75939.0768193936,
-      "required_pages_per_second_upper_estimate": 0.006582206712874446
+      "required_token_positions_per_second": 75935.77404819656,
+      "required_pages_per_second_upper_estimate": 0.006581920437037884
     },
     {
       "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": 9492.3846024242,
-      "required_pages_per_second_upper_estimate": 0.0037866541417042442
+      "required_token_positions_per_second": 9491.97175602457,
+      "required_pages_per_second_upper_estimate": 0.0037864894511028284
     },
     {
       "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": 18984.7692048484,
-      "required_pages_per_second_upper_estimate": 0.004391792235015304
+      "required_token_positions_per_second": 18983.94351204914,
+      "required_pages_per_second_upper_estimate": 0.004391601225517893
     },
     {
       "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": 28477.153807272596,
-      "required_pages_per_second_upper_estimate": 0.0017341092587837392
+      "required_token_positions_per_second": 28475.91526807371,
+      "required_pages_per_second_upper_estimate": 0.00173403383824464
     },
     {
       "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": 28477.153807272596,
-      "required_pages_per_second_upper_estimate": 0.002192250129890915
+      "required_token_positions_per_second": 28475.91526807371,
+      "required_pages_per_second_upper_estimate": 0.002192154783714889
     },
     {
       "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": 28477.153807272596,
-      "required_pages_per_second_upper_estimate": 0.35359094338344604
+      "required_token_positions_per_second": 28475.91526807371,
+      "required_pages_per_second_upper_estimate": 0.3535755648717199
     }
   ],
   "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": 0.11826667167957794,
+    "producer_batch_seconds": 0.29451535240630733,
     "producer_p95_seconds": 0.33940272399922833,
-    "gpu_update_seconds": 0.9708835435983478,
-    "producer_batches_per_second": 8.455467510824336,
-    "gpu_batches_per_second": 1.0299896487005424,
-    "producer_headroom": 8.209274259689835,
+    "gpu_update_seconds": 0.970925771471095,
+    "producer_batches_per_second": 3.3954087344839685,
+    "gpu_batches_per_second": 1.0299448519991938,
+    "producer_headroom": 3.2966898449885416,
     "sustainable": true,
-    "buffer_seconds": 60.194779703097566,
+    "buffer_seconds": 60.197397831207894,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 487267.07296152326,
+    "generation_bytes_per_second": 433163.44182516914,
     "upload_bytes_per_second": 3508235.566018321,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.55617086061739,
+    "estimated_full_shards_per_hour": 92.94679108682924,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -258,3 +258,4 @@
 {"step": 65800, "loss": 0.6213799938559532, "lr": 0.0003, "grad_norm": 0.13563732939225456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9630064669, "tokens_per_second": 148455.73038975097, "data_wait_s": 1.2536998838186264e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 65900, "loss": 0.6381329819560051, "lr": 0.0003, "grad_norm": 0.2403724516363505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9644776326, "tokens_per_second": 150310.7531065863, "data_wait_s": 1.0716998076532036e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 66000, "loss": 0.6565835401415825, "lr": 0.0003, "grad_norm": 0.1419560230238423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9659487014, "tokens_per_second": 149725.03250765483, "data_wait_s": 1.1422998795751482e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 66100, "loss": 0.662840723991394, "lr": 0.0003, "grad_norm": 0.13320195560030293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9674131831, "tokens_per_second": 149037.0426219479, "data_wait_s": 1.156200232799165e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e9f13e54ba5fdbf9580a53ed41646d8b28af57559adbb7f866f3aebed150545
-size 554760
+oid sha256:097febb999722a33f2b8e048178b17b4109f319b886f4c7dc50949f8b18cc6ee
+size 556878