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": 76092.97803062593,
-      "required_pages_per_second_upper_estimate": 0.006595546479805012
+      "required_token_positions_per_second": 76024.17600051314,
+      "required_pages_per_second_upper_estimate": 0.006589582894211986
     },
     {
       "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": 9511.622253828242,
-      "required_pages_per_second_upper_estimate": 0.003794328328477837
+      "required_token_positions_per_second": 9503.022000064142,
+      "required_pages_per_second_upper_estimate": 0.0037908975586660853
     },
     {
       "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": 19023.244507656484,
-      "required_pages_per_second_upper_estimate": 0.004400692819177753
+      "required_token_positions_per_second": 19006.044000128284,
+      "required_pages_per_second_upper_estimate": 0.004396713784479697
     },
     {
       "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": 28534.866761484725,
-      "required_pages_per_second_upper_estimate": 0.0017376236748987883
+      "required_token_positions_per_second": 28509.06600019243,
+      "required_pages_per_second_upper_estimate": 0.0017360525439022204
     },
     {
       "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": 28534.866761484725,
-      "required_pages_per_second_upper_estimate": 0.002196693032865847
+      "required_token_positions_per_second": 28509.06600019243,
+      "required_pages_per_second_upper_estimate": 0.002194706818840488
     },
     {
       "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": 28534.866761484725,
-      "required_pages_per_second_upper_estimate": 0.35430754512192814
+      "required_token_positions_per_second": 28509.06600019243,
+      "required_pages_per_second_upper_estimate": 0.35398718601627116
     }
   ],
   "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": 0.13273767969531036,
-    "producer_p95_seconds": 0.23366351900040172,
-    "gpu_update_seconds": 0.9689198912720425,
-    "producer_batches_per_second": 7.533655871455843,
-    "gpu_batches_per_second": 1.0320770674726825,
-    "producer_headroom": 7.29950902785198,
+    "producer_batch_seconds": 0.09048123760162241,
+    "producer_p95_seconds": 0.1997182420000172,
+    "gpu_update_seconds": 0.9697967656959854,
+    "producer_batches_per_second": 11.05201505314146,
+    "gpu_batches_per_second": 1.0311438802152932,
+    "producer_headroom": 10.718208452959932,
     "sustainable": true,
-    "buffer_seconds": 60.07303325886663,
+    "buffer_seconds": 60.127399473151094,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313800.66135605704,
+    "generation_bytes_per_second": 314248.84728296526,
     "upload_bytes_per_second": 5212065.682698314,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.33431702147755,
+    "estimated_full_shards_per_hour": 67.43048728815764,
     "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

@@ -97,3 +97,4 @@
 {"step": 164700, "loss": 0.6369891166687012, "lr": 0.0003, "grad_norm": 0.1658915428984174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24104947467, "tokens_per_second": 150312.88167741676, "data_wait_s": 1.4675999409519136e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 164800, "loss": 0.61604193598032, "lr": 0.0003, "grad_norm": 0.13755435102882466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24119541479, "tokens_per_second": 150061.13550155467, "data_wait_s": 1.4462000763160177e-05, "gpu_peak_gb": 85.46427917480469}
 {"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}

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 sha256:627dc5bb51786e7af1b36b383575fa9b294a78ac600028b469c2572499680e3b
-size 210630
+oid sha256:4cf0a3ccbbeb2b8035a7bb856999aa1f965faa8413073ce99be62d0ac80bb46b
+size 212748