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": 74586.64714375514,
-      "required_pages_per_second_upper_estimate": 0.00646498153629178
+      "required_token_positions_per_second": 75454.20361404722,
+      "required_pages_per_second_upper_estimate": 0.006540179132334926
     },
     {
       "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": 9323.330892969392,
-      "required_pages_per_second_upper_estimate": 0.003719216089424522
+      "required_token_positions_per_second": 9431.775451755902,
+      "required_pages_per_second_upper_estimate": 0.003762476245315104
     },
     {
       "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": 18646.661785938784,
-      "required_pages_per_second_upper_estimate": 0.004313577034137904
+      "required_token_positions_per_second": 18863.550903511805,
+      "required_pages_per_second_upper_estimate": 0.004363750514370324
     },
     {
       "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": 27969.992678908176,
-      "required_pages_per_second_upper_estimate": 0.0017032258069351438
+      "required_token_positions_per_second": 28295.326355267705,
+      "required_pages_per_second_upper_estimate": 0.0017230369209315548
     },
     {
       "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": 27969.992678908176,
-      "required_pages_per_second_upper_estimate": 0.0021532074623175643
+      "required_token_positions_per_second": 28295.326355267705,
+      "required_pages_per_second_upper_estimate": 0.002178252549304977
     },
     {
       "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": 27969.992678908176,
-      "required_pages_per_second_upper_estimate": 0.3472936995282687
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28295.326355267705,
+      "required_pages_per_second_upper_estimate": 0.3513332549668811
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.0928275197810251,
-    "producer_p95_seconds": 0.29370976599966525,
-    "gpu_update_seconds": 0.9884879240904848,
-    "producer_batches_per_second": 10.772667441281893,
-    "gpu_batches_per_second": 1.0116461472406024,
-    "producer_headroom": 10.648651675949893,
+    "producer_batch_seconds": 0.19491213583580702,
+    "producer_p95_seconds": 0.324173631001031,
+    "gpu_update_seconds": 0.9771224990607965,
+    "producer_batches_per_second": 5.130516864493214,
+    "gpu_batches_per_second": 1.0234131349561526,
+    "producer_headroom": 5.0131434601071705,
     "sustainable": true,
-    "buffer_seconds": 60.29776336951957,
+    "buffer_seconds": 60.58159494176938,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 446785.3179622072,
-    "upload_bytes_per_second": 4890778.764759392,
+    "generation_bytes_per_second": 465607.78508411674,
+    "upload_bytes_per_second": 4949429.028545004,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.86972860479032,
+    "estimated_full_shards_per_hour": 99.90859188454272,
     "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": 1.064075487000082,
-      "bytes": 6211136,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.7981215300023905,
       "bytes": 910382,
@@ -562,6 +556,12 @@
       "bytes": 5980251,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.218213135998667,
+      "bytes": 11903355,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -258,3 +258,4 @@
 {"step": 380800, "loss": 0.5672803446650505, "lr": 0.0003, "grad_norm": 0.16266216004883965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55737586324, "tokens_per_second": 149288.4222409075, "data_wait_s": 9.526000212645158e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 380900, "loss": 0.5989934783428907, "lr": 0.0003, "grad_norm": 0.17469817775212043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55752269287, "tokens_per_second": 149831.95597796407, "data_wait_s": 8.164999599102885e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 381000, "loss": 0.6876830719411373, "lr": 0.0003, "grad_norm": 0.23661556788380597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55766968013, "tokens_per_second": 148563.62394205033, "data_wait_s": 1.4056000509299338e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 381100, "loss": 0.6193846669048071, "lr": 0.0003, "grad_norm": 0.1722070226718004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55781639790, "tokens_per_second": 149361.8708315675, "data_wait_s": 2.0012997993035242e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3e0c1ae33efd63fad2283d8751e9c25a81e1456ef890a31e66048ad6cd54149
-size 554760
+oid sha256:1d6820d55520b3992bdef08a8988a8c2088e44752137c6948084847cea78e116
+size 556878