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": 73770.39557963757,
-      "required_pages_per_second_upper_estimate": 0.006394230919484744
+      "required_token_positions_per_second": 75030.93581121553,
+      "required_pages_per_second_upper_estimate": 0.006503491351947914
     },
     {
       "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": 9221.299447454696,
-      "required_pages_per_second_upper_estimate": 0.003678514220302655
+      "required_token_positions_per_second": 9378.866976401941,
+      "required_pages_per_second_upper_estimate": 0.0037413702634442082
     },
     {
       "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": 18442.59889490939,
-      "required_pages_per_second_upper_estimate": 0.004266370675682475
+      "required_token_positions_per_second": 18757.733952803883,
+      "required_pages_per_second_upper_estimate": 0.004339271625138772
     },
     {
       "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": 27663.898342364086,
-      "required_pages_per_second_upper_estimate": 0.001684586267792478
+      "required_token_positions_per_second": 28136.600929205826,
+      "required_pages_per_second_upper_estimate": 0.0017133713752523372
     },
     {
       "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": 27663.898342364086,
-      "required_pages_per_second_upper_estimate": 0.002129643472967046
+      "required_token_positions_per_second": 28136.600929205826,
+      "required_pages_per_second_upper_estimate": 0.0021660334266266324
     },
     {
       "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": 27663.898342364086,
-      "required_pages_per_second_upper_estimate": 0.34349303230023576
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28136.600929205826,
+      "required_pages_per_second_upper_estimate": 0.34936241639502125
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.12163342836788615,
-    "producer_p95_seconds": 0.39560157399682794,
-    "gpu_update_seconds": 0.9994253036152992,
-    "producer_batches_per_second": 8.221424105349163,
-    "gpu_batches_per_second": 1.0005750268505529,
-    "producer_headroom": 8.216699282638729,
+    "producer_batch_seconds": 0.36547426057853727,
+    "producer_p95_seconds": 0.40041848700639093,
+    "gpu_update_seconds": 0.9826346853184154,
+    "producer_batches_per_second": 2.7361708001461533,
+    "gpu_batches_per_second": 1.01767219796028,
+    "producer_headroom": 2.688656333179052,
     "sustainable": true,
-    "buffer_seconds": 60.964943520533254,
+    "buffer_seconds": 60.92335048974175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398448.20070988534,
-    "upload_bytes_per_second": 5705659.8649030505,
+    "generation_bytes_per_second": 358993.69832694693,
+    "upload_bytes_per_second": 5727892.134559002,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.49770847294255,
+    "estimated_full_shards_per_hour": 77.0316907153731,
     "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": 1.237680772999738,
-      "bytes": 6534213,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.075308318999305,
       "bytes": 6271877,
@@ -562,6 +556,12 @@
       "bytes": 2238236,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9348791060037911,
+      "bytes": 6426893,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -367,3 +367,4 @@
 {"step": 231700, "loss": 0.583973566070199, "lr": 0.0003, "grad_norm": 0.16444283742634583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33914520636, "tokens_per_second": 148426.13955550146, "data_wait_s": 1.5500998415518552e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 231800, "loss": 0.6330943591892719, "lr": 0.0003, "grad_norm": 0.15224543043698488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33929256447, "tokens_per_second": 149345.46635311778, "data_wait_s": 9.500006854068488e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 231900, "loss": 0.6245825625956059, "lr": 0.0003, "grad_norm": 0.17319920183125675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33944115985, "tokens_per_second": 151101.81390666304, "data_wait_s": 1.004899968393147e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 232000, "loss": 0.6067023389041424, "lr": 0.0003, "grad_norm": 0.1522415591060405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33958678080, "tokens_per_second": 148290.96934798567, "data_wait_s": 1.2838994734920561e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d69b174a98fac534c7df6d37d39a75098c2309b4f671b39affe5606936548c1
-size 787188
+oid sha256:1bcc0b50ad3701809a5d1110fe0db26685cdebdfa13df15659f39758ec37ef07
+size 789306