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": 76197.02034728209,
-      "required_pages_per_second_upper_estimate": 0.006604564604119926
+      "required_token_positions_per_second": 76098.14129398759,
+      "required_pages_per_second_upper_estimate": 0.006595994018387028
     },
     {
       "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": 9524.62754341026,
-      "required_pages_per_second_upper_estimate": 0.0037995163329385116
+      "required_token_positions_per_second": 9512.267661748449,
+      "required_pages_per_second_upper_estimate": 0.003794585791346916
     },
     {
       "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": 19049.25508682052,
-      "required_pages_per_second_upper_estimate": 0.004406709908896795
+      "required_token_positions_per_second": 19024.535323496897,
+      "required_pages_per_second_upper_estimate": 0.004400991426704849
     },
     {
       "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": 28573.882630230786,
-      "required_pages_per_second_upper_estimate": 0.0017399995366049802
+      "required_token_positions_per_second": 28536.802985245344,
+      "required_pages_per_second_upper_estimate": 0.0017377415807672291
     },
     {
       "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": 28573.882630230786,
-      "required_pages_per_second_upper_estimate": 0.002199696582444756
+      "required_token_positions_per_second": 28536.802985245344,
+      "required_pages_per_second_upper_estimate": 0.00219684208873075
     },
     {
       "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": 28573.882630230786,
-      "required_pages_per_second_upper_estimate": 0.35479199163404135
+      "required_token_positions_per_second": 28536.802985245344,
+      "required_pages_per_second_upper_estimate": 0.35433158654091096
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10045125473425287,
+    "producer_batch_seconds": 0.0873455412186388,
     "producer_p95_seconds": 0.3286219030014763,
-    "gpu_update_seconds": 0.967596891111633,
-    "producer_batches_per_second": 9.95507724264404,
-    "gpu_batches_per_second": 1.0334882317068426,
-    "producer_headroom": 9.63250179075854,
+    "gpu_update_seconds": 0.9688541500004436,
+    "producer_batches_per_second": 11.44878131210902,
+    "gpu_batches_per_second": 1.0321470987140242,
+    "producer_headroom": 11.092199286684348,
     "sustainable": true,
-    "buffer_seconds": 60.95860414003288,
+    "buffer_seconds": 60.068957300027506,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287035.7895652393,
-    "upload_bytes_per_second": 4462140.27604122,
+    "generation_bytes_per_second": 315061.3361149651,
+    "upload_bytes_per_second": 4455221.885658993,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.591198589495505,
+    "estimated_full_shards_per_hour": 67.6048284777328,
     "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": [
-    {
-      "seconds": 1.182673250001244,
-      "bytes": 5984869,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.7554075879997981,
       "bytes": 949451,
@@ -562,6 +556,12 @@
       "bytes": 2231089,
       "pages": 24,
       "shards": 2
+    },
+    {
+      "seconds": 1.3213498320001236,
+      "bytes": 5953362,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -300,3 +300,4 @@
 {"step": 185000, "loss": 0.5817938577383757, "lr": 0.0003, "grad_norm": 0.151997777867797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27076235668, "tokens_per_second": 149048.55473883892, "data_wait_s": 9.453000529902056e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 185100, "loss": 0.616595970466733, "lr": 0.0003, "grad_norm": 0.13262788098500142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27090932419, "tokens_per_second": 130173.93343251415, "data_wait_s": 8.660997991682962e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 185200, "loss": 0.6223879046738148, "lr": 0.0003, "grad_norm": 0.13899563103482293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27105600913, "tokens_per_second": 150016.95005225413, "data_wait_s": 7.956998160807416e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 185300, "loss": 0.578156553208828, "lr": 0.0003, "grad_norm": 0.1414885166748943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27120082487, "tokens_per_second": 148110.79503216466, "data_wait_s": 9.035000402946025e-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:fac6eb53e085a2dd6c46a5e36633c7b1deae92f450c7ae937f7448c371e208f1
-size 644499
+oid sha256:86b4b0b925218522cc2fe4e3a917394d1528ecabd4943e070f0b7c6165f3fb7a
+size 646617