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": 74807.03257400253,
-      "required_pages_per_second_upper_estimate": 0.006484083986823858
+      "required_token_positions_per_second": 75902.60727423328,
+      "required_pages_per_second_upper_estimate": 0.006579045625131167
     },
     {
       "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": 9350.879071750316,
-      "required_pages_per_second_upper_estimate": 0.00373020546982221
+      "required_token_positions_per_second": 9487.82590927916,
+      "required_pages_per_second_upper_estimate": 0.0037848356108501516
     },
     {
       "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": 18701.758143500632,
-      "required_pages_per_second_upper_estimate": 0.004326322606797063
+      "required_token_positions_per_second": 18975.65181855832,
+      "required_pages_per_second_upper_estimate": 0.004389683088157613
     },
     {
       "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": 28052.637215250947,
-      "required_pages_per_second_upper_estimate": 0.0017082584255960467
+      "required_token_positions_per_second": 28463.47772783748,
+      "required_pages_per_second_upper_estimate": 0.0017332764572989838
     },
     {
       "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": 28052.637215250947,
-      "required_pages_per_second_upper_estimate": 0.00215956966749995
+      "required_token_positions_per_second": 28463.47772783748,
+      "required_pages_per_second_upper_estimate": 0.002191197307438189
     },
     {
       "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": 28052.637215250947,
-      "required_pages_per_second_upper_estimate": 0.3483198680761755
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28463.47772783748,
+      "required_pages_per_second_upper_estimate": 0.3534211322477554
     }
   ],
   "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": 0.1016963352812752,
-    "producer_p95_seconds": 0.22090510500129312,
-    "gpu_update_seconds": 0.9855757869698267,
-    "producer_batches_per_second": 9.833196026525105,
-    "gpu_batches_per_second": 1.0146353159451298,
-    "producer_headroom": 9.691359912271052,
+    "producer_batch_seconds": 0.09810307842184329,
+    "producer_p95_seconds": 0.20878076200096984,
+    "gpu_update_seconds": 0.9713500319379478,
+    "producer_batches_per_second": 10.193360046256647,
+    "gpu_batches_per_second": 1.0294949988367144,
+    "producer_headroom": 9.901320606486395,
     "sustainable": true,
-    "buffer_seconds": 60.12012300515943,
+    "buffer_seconds": 60.22370198015276,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272959.5530671881,
-    "upload_bytes_per_second": 3706594.1854007556,
+    "generation_bytes_per_second": 281094.2113594621,
+    "upload_bytes_per_second": 3730402.716901991,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.570765915028886,
+    "estimated_full_shards_per_hour": 60.31627421939752,
     "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": 1.3083305300001484,
-      "bytes": 5580567,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1845111140000881,
       "bytes": 11699180,
@@ -562,6 +556,12 @@
       "bytes": 6655300,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0988184660000115,
+      "bytes": 7486365,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -90,3 +90,4 @@
 {"step": 246100, "loss": 0.5597578007727861, "lr": 0.0003, "grad_norm": 0.13582123623624906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36022129649, "tokens_per_second": 149542.31262298633, "data_wait_s": 1.5158000678638928e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 246200, "loss": 0.6082983128726482, "lr": 0.0003, "grad_norm": 0.14485082765122037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36036857721, "tokens_per_second": 150080.03273420228, "data_wait_s": 9.61499972618185e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 246300, "loss": 0.5327975749969482, "lr": 0.0003, "grad_norm": 0.15585395626792478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36051369840, "tokens_per_second": 147922.12861020927, "data_wait_s": 1.4101000488153659e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 246400, "loss": 0.6401682272553444, "lr": 0.0003, "grad_norm": 0.14270779245222198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36065781690, "tokens_per_second": 146995.02602726058, "data_wait_s": 1.0323999958927743e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ed0f6a387b2d78cf5884645378e8a1c0431e114c58ae37debe7a89d813eae05d
-size 196587
+oid sha256:ffe9f08647c87fd5cbf86a03f6344cd833a23f55e466e74b3599081e1e47cbc3
+size 198705