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": 75833.98217678783,
-      "required_pages_per_second_upper_estimate": 0.006573097375613315
+      "required_token_positions_per_second": 75562.6393274024,
+      "required_pages_per_second_upper_estimate": 0.006549578065140751
     },
     {
       "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": 9479.247772098479,
-      "required_pages_per_second_upper_estimate": 0.0037814136636741976
+      "required_token_positions_per_second": 9445.3299159253,
+      "required_pages_per_second_upper_estimate": 0.0037678833237295758
     },
     {
       "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": 18958.495544196958,
-      "required_pages_per_second_upper_estimate": 0.004385714286024228
+      "required_token_positions_per_second": 18890.6598318506,
+      "required_pages_per_second_upper_estimate": 0.00437002168784066
     },
     {
       "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": 28437.743316295437,
-      "required_pages_per_second_upper_estimate": 0.001731709366654096
+      "required_token_positions_per_second": 28335.9897477759,
+      "required_pages_per_second_upper_estimate": 0.0017255131081909715
     },
     {
       "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": 28437.743316295437,
-      "required_pages_per_second_upper_estimate": 0.002189216201200272
+      "required_token_positions_per_second": 28335.9897477759,
+      "required_pages_per_second_upper_estimate": 0.0021813829298237327
     },
     {
       "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": 28437.743316295437,
-      "required_pages_per_second_upper_estimate": 0.3531015969839383
+      "required_token_positions_per_second": 28335.9897477759,
+      "required_pages_per_second_upper_estimate": 0.3518381582102127
     }
   ],
   "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": 0.05538229834378683,
-    "producer_p95_seconds": 0.1910304120001456,
-    "gpu_update_seconds": 0.9722290440731668,
-    "producer_batches_per_second": 18.056310949619284,
-    "gpu_batches_per_second": 1.0285642113822133,
-    "producer_headroom": 17.55486993403621,
+    "producer_batch_seconds": 0.11335125493740605,
+    "producer_p95_seconds": 0.1996027059994958,
+    "gpu_update_seconds": 0.9757202852661992,
+    "producer_batches_per_second": 8.822134351774157,
+    "gpu_batches_per_second": 1.0248838884467557,
+    "producer_headroom": 8.607935446369817,
     "sustainable": true,
-    "buffer_seconds": 60.27820073253634,
+    "buffer_seconds": 60.494657686504354,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298238.97390492645,
-    "upload_bytes_per_second": 4902214.036642064,
+    "generation_bytes_per_second": 298270.5979521755,
+    "upload_bytes_per_second": 4929764.27135805,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.995141152008486,
+    "estimated_full_shards_per_hour": 64.00192693637798,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 1181305,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.226145381000606,
+      "bytes": 7407097,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -74,3 +74,4 @@
 {"step": 162400, "loss": 0.5808803290128708, "lr": 0.0003, "grad_norm": 0.13058085458923926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23768096827, "tokens_per_second": 148903.8294180968, "data_wait_s": 1.1887999789905734e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 162500, "loss": 0.5652763843536377, "lr": 0.0003, "grad_norm": 0.1424294736923307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23782722685, "tokens_per_second": 149719.37303600585, "data_wait_s": 1.440800042473711e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 162600, "loss": 0.5990036465227604, "lr": 0.0003, "grad_norm": 0.12867135792371365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23797449426, "tokens_per_second": 149717.1590024211, "data_wait_s": 1.1140999959025066e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 162700, "loss": 0.5720186643302441, "lr": 0.0003, "grad_norm": 0.1345017926209548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23812088597, "tokens_per_second": 149943.13882640048, "data_wait_s": 1.2315998901613057e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93bea50d39f6051abc122ca8f5783d036e5f0c5c050302df1a1e144855420dc7
-size 161916
+oid sha256:3381f18581958167250e24e1646e68d9a2baea0d5de25f788d73f9e13101ef75
+size 164034