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": 76024.17600051314,
-      "required_pages_per_second_upper_estimate": 0.006589582894211986
+      "required_token_positions_per_second": 76303.2876998212,
+      "required_pages_per_second_upper_estimate": 0.006613775588905874
     },
     {
       "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": 9503.022000064142,
-      "required_pages_per_second_upper_estimate": 0.0037908975586660853
+      "required_token_positions_per_second": 9537.91096247765,
+      "required_pages_per_second_upper_estimate": 0.003804815287409307
     },
     {
       "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": 19006.044000128284,
-      "required_pages_per_second_upper_estimate": 0.004396713784479697
+      "required_token_positions_per_second": 19075.8219249553,
+      "required_pages_per_second_upper_estimate": 0.0044128556793388935
     },
     {
       "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": 28509.06600019243,
-      "required_pages_per_second_upper_estimate": 0.0017360525439022204
+      "required_token_positions_per_second": 28613.732887432947,
+      "required_pages_per_second_upper_estimate": 0.0017424262081904508
     },
     {
       "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": 28509.06600019243,
-      "required_pages_per_second_upper_estimate": 0.002194706818840488
+      "required_token_positions_per_second": 28613.732887432947,
+      "required_pages_per_second_upper_estimate": 0.0022027643655567505
     },
     {
       "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": 28509.06600019243,
-      "required_pages_per_second_upper_estimate": 0.35398718601627116
+      "required_token_positions_per_second": 28613.732887432947,
+      "required_pages_per_second_upper_estimate": 0.3552867984582608
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "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,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.127399473151094,
+    "buffer_seconds": 60.87370728077926,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314248.84728296526,
-    "upload_bytes_per_second": 5212065.682698314,
+    "generation_bytes_per_second": 315149.76968688704,
+    "upload_bytes_per_second": 5180020.596017086,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.43048728815764,
+    "estimated_full_shards_per_hour": 67.62380426363906,
     "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": [
@@ -520,6 +520,18 @@
       "bytes": 9934254,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9314350770000601,
+      "bytes": 2320145,
+      "pages": 27,
+      "shards": 2
+    },
+    {
+      "seconds": 1.1146210560000327,
+      "bytes": 5994268,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -98,3 +98,4 @@
 {"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}
+{"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}

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 sha256:98fc8b99abcdf3aaf184108fae2ab842a4c8122da22294e4cb4dc4a6ec477ecf
-size 213531
+oid sha256:f5efe01bfa80b9c18b0c36bf42945e39acb66f2b091d7d048f22b45ad97f938d
+size 215649