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": 74864.38749097098,
-      "required_pages_per_second_upper_estimate": 0.006489055365661975
+      "required_token_positions_per_second": 75451.13994058831,
+      "required_pages_per_second_upper_estimate": 0.006539913580884318
     },
     {
       "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": 9358.048436371373,
-      "required_pages_per_second_upper_estimate": 0.003733065436561103
+      "required_token_positions_per_second": 9431.39249257354,
+      "required_pages_per_second_upper_estimate": 0.003762323477171509
     },
     {
       "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": 18716.096872742746,
-      "required_pages_per_second_upper_estimate": 0.00432963961945421
+      "required_token_positions_per_second": 18862.78498514708,
+      "required_pages_per_second_upper_estimate": 0.004363573332636347
     },
     {
       "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": 28074.14530911412,
-      "required_pages_per_second_upper_estimate": 0.0017095681556680666
+      "required_token_positions_per_second": 28294.17747772062,
+      "required_pages_per_second_upper_estimate": 0.001722966960316632
     },
     {
       "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": 28074.14530911412,
-      "required_pages_per_second_upper_estimate": 0.002161225420103751
+      "required_token_positions_per_second": 28294.17747772062,
+      "required_pages_per_second_upper_estimate": 0.002178164105531106
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28074.14530911412,
-      "required_pages_per_second_upper_estimate": 0.3485869266189965
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28294.17747772062,
+      "required_pages_per_second_upper_estimate": 0.35131898975279213
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.44185244066405005,
-    "producer_p95_seconds": 0.20105442800013407,
-    "gpu_update_seconds": 0.9848207201173182,
-    "producer_batches_per_second": 2.2631989957939864,
-    "gpu_batches_per_second": 1.015413241793769,
-    "producer_headroom": 2.2288452648066253,
+    "producer_batch_seconds": 0.0996374236796953,
+    "producer_p95_seconds": 0.19691285000044445,
+    "gpu_update_seconds": 0.9771621748598477,
+    "producer_batches_per_second": 10.036389572001609,
+    "gpu_batches_per_second": 1.0233715812254276,
+    "producer_headroom": 9.807180261917788,
     "sustainable": true,
-    "buffer_seconds": 60.074063927156416,
+    "buffer_seconds": 60.584054841310554,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 464573.9473682442,
-    "upload_bytes_per_second": 3972031.611808549,
+    "generation_bytes_per_second": 482475.7310035794,
+    "upload_bytes_per_second": 4020639.445295186,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.68675437722678,
+    "estimated_full_shards_per_hour": 103.52806041317497,
     "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": 0.9309580040001038,
-      "bytes": 6008731,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.3192515309999635,
       "bytes": 6183122,
@@ -562,6 +556,12 @@
       "bytes": 11705717,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.197896118999779,
+      "bytes": 12261236,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -84,3 +84,4 @@
 {"step": 323400, "loss": 0.5881020314991474, "lr": 0.0003, "grad_norm": 0.14270557128536146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47336442704, "tokens_per_second": 149093.52640719488, "data_wait_s": 1.467500078433659e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 323500, "loss": 0.5915633793920279, "lr": 0.0003, "grad_norm": 0.15062822252435012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47351181869, "tokens_per_second": 149537.47764343666, "data_wait_s": 1.164499917649664e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 323600, "loss": 0.5999933667480946, "lr": 0.0003, "grad_norm": 0.17164955116625333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47365910119, "tokens_per_second": 150341.68643233628, "data_wait_s": 1.4494999049929902e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 323700, "loss": 0.6073876842856407, "lr": 0.0003, "grad_norm": 0.1726146494891655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47380602911, "tokens_per_second": 150101.78471536125, "data_wait_s": 1.389600038237404e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3dfd58a5c75d88210e195be24cacacad74745be44b494ffd92b86d6a523bb8fd
-size 183096
+oid sha256:4338c18f7c3faa80aceebe41b27ee88bba162d04b1d88343a79d45d68f19766f
+size 185214