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": 76303.2876998212,
-      "required_pages_per_second_upper_estimate": 0.006613775588905874
+      "required_token_positions_per_second": 76240.172578192,
+      "required_pages_per_second_upper_estimate": 0.006608304930126868
     },
     {
       "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": 9537.91096247765,
-      "required_pages_per_second_upper_estimate": 0.003804815287409307
+      "required_token_positions_per_second": 9530.021572274,
+      "required_pages_per_second_upper_estimate": 0.0038016680917001755
     },
     {
       "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": 19075.8219249553,
-      "required_pages_per_second_upper_estimate": 0.0044128556793388935
+      "required_token_positions_per_second": 19060.043144548,
+      "required_pages_per_second_upper_estimate": 0.004409205536188715
     },
     {
       "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": 28613.732887432947,
-      "required_pages_per_second_upper_estimate": 0.0017424262081904508
+      "required_token_positions_per_second": 28590.064716821995,
+      "required_pages_per_second_upper_estimate": 0.001740984940777538
     },
     {
       "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": 28613.732887432947,
-      "required_pages_per_second_upper_estimate": 0.0022027643655567505
+      "required_token_positions_per_second": 28590.064716821995,
+      "required_pages_per_second_upper_estimate": 0.00220094232426543
     },
     {
       "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": 28613.732887432947,
-      "required_pages_per_second_upper_estimate": 0.3552867984582608
+      "required_token_positions_per_second": 28590.064716821995,
+      "required_pages_per_second_upper_estimate": 0.35499291899154417
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08143183573437796,
-    "producer_p95_seconds": 0.20509806100017158,
-    "gpu_update_seconds": 0.9662493219171311,
-    "producer_batches_per_second": 12.280209465765875,
-    "gpu_batches_per_second": 1.0349295749216199,
-    "producer_headroom": 11.86574406929661,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.87370728077926,
+    "buffer_seconds": 60.924101335634084,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315149.76968688704,
+    "generation_bytes_per_second": 360164.63081734185,
     "upload_bytes_per_second": 5180020.596017086,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.62380426363906,
+    "estimated_full_shards_per_hour": 77.28294556989853,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -99,3 +99,4 @@
 {"step": 164900, "loss": 0.6301826126873493, "lr": 0.0003, "grad_norm": 0.16130835641863508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24134025368, "tokens_per_second": 147929.0721923293, "data_wait_s": 1.3327999113244005e-05, "gpu_peak_gb": 85.46427917480469}
 {"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}

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:f5efe01bfa80b9c18b0c36bf42945e39acb66f2b091d7d048f22b45ad97f938d
-size 215649
+oid sha256:b580903d69879d02b87bfd11c663efe8eb20aca1acc0f84b7a8ab8cbce8b7b62
+size 217767