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": 76127.35069493848,
-      "required_pages_per_second_upper_estimate": 0.006598525815230901
+      "required_token_positions_per_second": 73957.38992688672,
+      "required_pages_per_second_upper_estimate": 0.006410439115571462
     },
     {
       "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": 9515.91883686731,
-      "required_pages_per_second_upper_estimate": 0.003796042299691762
+      "required_token_positions_per_second": 9244.67374086084,
+      "required_pages_per_second_upper_estimate": 0.003687838575419196
     },
     {
       "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": 19031.83767373462,
-      "required_pages_per_second_upper_estimate": 0.004402680696915372
+      "required_token_positions_per_second": 18489.34748172168,
+      "required_pages_per_second_upper_estimate": 0.00427718513849447
     },
     {
       "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": 28547.756510601925,
-      "required_pages_per_second_upper_estimate": 0.0017384085929927388
+      "required_token_positions_per_second": 27734.021222582523,
+      "required_pages_per_second_upper_estimate": 0.0016888563832914616
     },
     {
       "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": 28547.756510601925,
-      "required_pages_per_second_upper_estimate": 0.0021976853214339983
+      "required_token_positions_per_second": 27734.021222582523,
+      "required_pages_per_second_upper_estimate": 0.0021350417263988172
     },
     {
       "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": 28547.756510601925,
-      "required_pages_per_second_upper_estimate": 0.3544675926667485
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27734.021222582523,
+      "required_pages_per_second_upper_estimate": 0.34436372378636554
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 688946,
-    "producer_batch_seconds": 0.12316699090627026,
-    "producer_p95_seconds": 0.3602621399986674,
-    "gpu_update_seconds": 0.9684824091074274,
-    "producer_batches_per_second": 8.11905846397593,
-    "gpu_batches_per_second": 1.0325432765698035,
-    "producer_headroom": 7.863165300875458,
+    "producer_batch_seconds": 0.45205453984374344,
+    "producer_p95_seconds": 0.37967734199992265,
+    "gpu_update_seconds": 0.9968983501565767,
+    "producer_batches_per_second": 2.212122458377829,
+    "gpu_batches_per_second": 1.0031113000066016,
+    "producer_headroom": 2.2052612291011684,
     "sustainable": true,
-    "buffer_seconds": 60.0459093646605,
+    "buffer_seconds": 60.810799359551176,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 543837.8009776565,
+    "generation_bytes_per_second": 480005.77806478593,
     "upload_bytes_per_second": 5132226.018885838,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.69493219373008,
+    "estimated_full_shards_per_hour": 102.9980660100716,
     "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": [

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -319,3 +319,4 @@
 {"step": 146900, "loss": 0.6355623006820679, "lr": 0.0003, "grad_norm": 0.1646902456067799, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21498590125, "tokens_per_second": 150710.4243415569, "data_wait_s": 1.0867999662877992e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 147000, "loss": 0.5945109017193317, "lr": 0.0003, "grad_norm": 0.13763741233161803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21513190824, "tokens_per_second": 149468.14629271862, "data_wait_s": 9.603001672076061e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 147100, "loss": 0.5724699255079031, "lr": 0.0003, "grad_norm": 0.1468198003576462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21527672095, "tokens_per_second": 147824.88346666755, "data_wait_s": 1.1464999261079356e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 147200, "loss": 0.6118672527372837, "lr": 0.0003, "grad_norm": 0.1628369852678089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21542244994, "tokens_per_second": 149631.85286732164, "data_wait_s": 9.551000403007492e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a6170d8182f8215502fe6a3845b239227ab9c191a133e03325ece410447ffee
-size 684741
+oid sha256:63162ede0a03fbbeb39adc1318c0c90d8ed44e8cbc0d4ad027fd085b5906ad42
+size 686859