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": 75033.46580332551,
-      "required_pages_per_second_upper_estimate": 0.006503710645251801
+      "required_token_positions_per_second": 73936.63402372906,
+      "required_pages_per_second_upper_estimate": 0.006408640046491111
     },
     {
       "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": 9379.183225415689,
-      "required_pages_per_second_upper_estimate": 0.003741496419904136
+      "required_token_positions_per_second": 9242.079252966132,
+      "required_pages_per_second_upper_estimate": 0.0036868035954069458
     },
     {
       "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": 18758.366450831378,
-      "required_pages_per_second_upper_estimate": 0.004339417942425848
+      "required_token_positions_per_second": 18484.158505932264,
+      "required_pages_per_second_upper_estimate": 0.004275984760268444
     },
     {
       "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": 28137.54967624707,
-      "required_pages_per_second_upper_estimate": 0.001713429148969458
+      "required_token_positions_per_second": 27726.237758898395,
+      "required_pages_per_second_upper_estimate": 0.001688382411189776
     },
     {
       "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": 28137.54967624707,
-      "required_pages_per_second_upper_estimate": 0.002166106463800167
+      "required_token_positions_per_second": 27726.237758898395,
+      "required_pages_per_second_upper_estimate": 0.0021344425338184064
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28137.54967624707,
-      "required_pages_per_second_upper_estimate": 0.3493741966580214
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27726.237758898395,
+      "required_pages_per_second_upper_estimate": 0.3442670792169859
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09271371146876106,
-    "producer_p95_seconds": 0.20319681000000855,
-    "gpu_update_seconds": 0.98260155266255,
-    "producer_batches_per_second": 10.78589114984292,
-    "gpu_batches_per_second": 1.0177065131744454,
-    "producer_headroom": 10.59823339068491,
+    "producer_batch_seconds": 0.4637958972655838,
+    "producer_p95_seconds": 0.23369430899947474,
+    "gpu_update_seconds": 0.9971782050064371,
+    "producer_batches_per_second": 2.1561208408606714,
+    "gpu_batches_per_second": 1.0028297800527486,
+    "producer_headroom": 2.1500367098664137,
     "sustainable": true,
-    "buffer_seconds": 60.9212962650781,
+    "buffer_seconds": 60.82787050539266,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319916.6696652353,
-    "upload_bytes_per_second": 4045039.515985068,
+    "generation_bytes_per_second": 287049.42554048105,
+    "upload_bytes_per_second": 4032983.6890908717,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.64667003124042,
+    "estimated_full_shards_per_hour": 61.59412455235313,
     "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": 0.8416516239994962,
-      "bytes": 2569575,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.165861663999749,
       "bytes": 11609605,
@@ -562,6 +556,12 @@
       "bytes": 11651730,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0898228620008013,
+      "bytes": 2320314,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -123,3 +123,4 @@
 {"step": 327300, "loss": 0.5769318398088217, "lr": 0.0003, "grad_norm": 0.15720138966766922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47907710928, "tokens_per_second": 148577.17979066292, "data_wait_s": 1.4534000001731329e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 327400, "loss": 0.6022498216480017, "lr": 0.0003, "grad_norm": 0.1598802955628392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47922314253, "tokens_per_second": 148855.92447491825, "data_wait_s": 1.4339999324874952e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 327500, "loss": 0.6007689274847507, "lr": 0.0003, "grad_norm": 0.13406897571219664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47936968395, "tokens_per_second": 137695.06942940524, "data_wait_s": 8.6880008893786e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 327600, "loss": 0.6296123526990414, "lr": 0.0003, "grad_norm": 0.15900981263965358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47951661711, "tokens_per_second": 149924.10939036196, "data_wait_s": 1.2664000678341836e-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 sha256:90153b64de14f3e9f74a6ca276ca2e6bfe2237b069a3dd5e6f89db0ad544881c
-size 266481
+oid sha256:556c9da8f93e26ea984b2a6b6cdf9e89ad7e3cd85e1d0fbd2a7f207d950f7db7
+size 268599