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": 75477.41239052585,
-      "required_pages_per_second_upper_estimate": 0.006542190810257987
+      "required_token_positions_per_second": 76258.97730875321,
+      "required_pages_per_second_upper_estimate": 0.0066099348765117595
     },
     {
       "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": 9434.67654881573,
-      "required_pages_per_second_upper_estimate": 0.003763633536307536
+      "required_token_positions_per_second": 9532.372163594151,
+      "required_pages_per_second_upper_estimate": 0.003802605777722256
     },
     {
       "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": 18869.35309763146,
-      "required_pages_per_second_upper_estimate": 0.004365092749864777
+      "required_token_positions_per_second": 19064.744327188302,
+      "required_pages_per_second_upper_estimate": 0.004410293072054561
     },
     {
       "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": 28304.02964644719,
-      "required_pages_per_second_upper_estimate": 0.0017235669057017457
+      "required_token_positions_per_second": 28597.116490782453,
+      "required_pages_per_second_upper_estimate": 0.0017414143568139312
     },
     {
       "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": 28304.02964644719,
-      "required_pages_per_second_upper_estimate": 0.0021789225527522257
+      "required_token_positions_per_second": 28597.116490782453,
+      "required_pages_per_second_upper_estimate": 0.002201485189345466
     },
     {
       "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": 28304.02964644719,
-      "required_pages_per_second_upper_estimate": 0.3514413207152885
+      "required_token_positions_per_second": 28597.116490782453,
+      "required_pages_per_second_upper_estimate": 0.3550804784233638
     }
   ],
   "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": 0.1281861842968226,
-    "producer_p95_seconds": 0.3227653569992981,
-    "gpu_update_seconds": 0.9768220407255849,
-    "producer_batches_per_second": 7.801152717709746,
-    "gpu_batches_per_second": 1.0237279241336512,
-    "producer_headroom": 7.620337917725177,
+    "producer_batch_seconds": 0.10864058887523242,
+    "producer_p95_seconds": 0.33304966099967714,
+    "gpu_update_seconds": 0.9668107625085776,
+    "producer_batches_per_second": 9.204662919753163,
+    "gpu_batches_per_second": 1.0343285767788792,
+    "producer_headroom": 8.899167176080987,
     "sustainable": true,
-    "buffer_seconds": 60.562966524986265,
+    "buffer_seconds": 60.909078038040384,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 463414.0546685952,
+    "generation_bytes_per_second": 462967.1187985985,
     "upload_bytes_per_second": 4551169.508737466,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.43786840480224,
+    "estimated_full_shards_per_hour": 99.34196637123551,
     "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

@@ -324,3 +324,4 @@
 {"step": 187400, "loss": 0.5586118642240763, "lr": 0.0003, "grad_norm": 0.1396880538457402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27427035330, "tokens_per_second": 149454.32112442594, "data_wait_s": 7.309998181881383e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 187500, "loss": 0.6721691600978374, "lr": 0.0003, "grad_norm": 0.17449702625152358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27441551250, "tokens_per_second": 148510.3078524755, "data_wait_s": 1.0808005754370242e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 187600, "loss": 0.569325465708971, "lr": 0.0003, "grad_norm": 0.14907051043626307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27456177803, "tokens_per_second": 151009.2038367556, "data_wait_s": 6.381997081916779e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 187700, "loss": 0.5929309464991093, "lr": 0.0003, "grad_norm": 0.14143837608252244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27471036553, "tokens_per_second": 152747.38312879467, "data_wait_s": 7.114002073649317e-06, "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 sha256:5531b7d83060ed9b6f44ebc5c7f40400f74670df8a410b2708173af13ebccca2
-size 695331
+oid sha256:6136c0525c76a4835590056d7f063d3489ea11df456db34461c60a32f2cfaf9c
+size 697449