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": 76240.172578192,
-      "required_pages_per_second_upper_estimate": 0.006608304930126868
+      "required_token_positions_per_second": 74867.70268204794,
+      "required_pages_per_second_upper_estimate": 0.0064893427180222505
     },
     {
       "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": 9530.021572274,
-      "required_pages_per_second_upper_estimate": 0.0038016680917001755
+      "required_token_positions_per_second": 9358.462835255992,
+      "required_pages_per_second_upper_estimate": 0.003733230746471993
     },
     {
       "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": 19060.043144548,
-      "required_pages_per_second_upper_estimate": 0.004409205536188715
+      "required_token_positions_per_second": 18716.925670511984,
+      "required_pages_per_second_upper_estimate": 0.00432983134723178
     },
     {
       "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": 28590.064716821995,
-      "required_pages_per_second_upper_estimate": 0.001740984940777538
+      "required_token_positions_per_second": 28075.38850576798,
+      "required_pages_per_second_upper_estimate": 0.001709643859821738
     },
     {
       "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": 28590.064716821995,
-      "required_pages_per_second_upper_estimate": 0.00220094232426543
+      "required_token_positions_per_second": 28075.38850576798,
+      "required_pages_per_second_upper_estimate": 0.0021613211248235537
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28590.064716821995,
-      "required_pages_per_second_upper_estimate": 0.35499291899154417
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28075.38850576798,
+      "required_pages_per_second_upper_estimate": 0.3486023629607259
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09059537054686473,
-    "producer_p95_seconds": 0.23538994199952867,
-    "gpu_update_seconds": 0.9670492275497474,
-    "producer_batches_per_second": 11.038091615097516,
-    "gpu_batches_per_second": 1.034073521297092,
-    "producer_headroom": 10.674377970003396,
+    "producer_batch_seconds": 0.1896823422735281,
+    "producer_p95_seconds": 0.21498337400043965,
+    "gpu_update_seconds": 0.9847771116086186,
+    "producer_batches_per_second": 5.2719720139156,
+    "gpu_batches_per_second": 1.01545820695052,
+    "producer_headroom": 5.191717372345276,
     "sustainable": true,
-    "buffer_seconds": 60.924101335634084,
+    "buffer_seconds": 60.07140380812574,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360164.63081734185,
-    "upload_bytes_per_second": 5180020.596017086,
+    "generation_bytes_per_second": 325293.5297610107,
+    "upload_bytes_per_second": 5133600.991084645,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.28294556989853,
+    "estimated_full_shards_per_hour": 69.80041904089681,
     "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": [
@@ -532,6 +532,12 @@
       "bytes": 5994268,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.7972013050002715,
+      "bytes": 688709,
+      "pages": 19,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -100,3 +100,4 @@
 {"step": 165000, "loss": 0.6499907895922661, "lr": 0.0003, "grad_norm": 0.14361916422261714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24148524258, "tokens_per_second": 148431.67601297708, "data_wait_s": 1.0272999134031124e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 165100, "loss": 0.61792853474617, "lr": 0.0003, "grad_norm": 0.14505391141285598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24163166765, "tokens_per_second": 130473.82964356792, "data_wait_s": 1.2906999472761527e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 165200, "loss": 0.5766500700265169, "lr": 0.0003, "grad_norm": 0.14344728605592386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24177823023, "tokens_per_second": 129855.91090325394, "data_wait_s": 1.2578000678331591e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 165300, "loss": 0.6171917375177145, "lr": 0.0003, "grad_norm": 0.15698313541763897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24192426010, "tokens_per_second": 149378.49429265875, "data_wait_s": 1.5203000657493249e-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:b580903d69879d02b87bfd11c663efe8eb20aca1acc0f84b7a8ab8cbce8b7b62
-size 217767
+oid sha256:d61f207bb44ee03364246a7ae2d7a019d47598adce8dfb5709eebcc2349f5246
+size 219885