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": 75656.06942469298,
-      "required_pages_per_second_upper_estimate": 0.006557676349177481
+      "required_token_positions_per_second": 74635.62718118774,
+      "required_pages_per_second_upper_estimate": 0.006469227001797668
     },
     {
       "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": 9457.008678086622,
-      "required_pages_per_second_upper_estimate": 0.0037725421565687816
+      "required_token_positions_per_second": 9329.453397648467,
+      "required_pages_per_second_upper_estimate": 0.0037216584480806074
     },
     {
       "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": 18914.017356173244,
-      "required_pages_per_second_upper_estimate": 0.004375425040014316
+      "required_token_positions_per_second": 18658.906795296934,
+      "required_pages_per_second_upper_estimate": 0.004316409701548107
     },
     {
       "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": 28371.026034259867,
-      "required_pages_per_second_upper_estimate": 0.0017276466342166564
+      "required_token_positions_per_second": 27988.360192945398,
+      "required_pages_per_second_upper_estimate": 0.0017043442921731119
     },
     {
       "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": 28371.026034259867,
-      "required_pages_per_second_upper_estimate": 0.0021840801201439167
+      "required_token_positions_per_second": 27988.360192945398,
+      "required_pages_per_second_upper_estimate": 0.0021546214443926826
     },
     {
       "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": 28371.026034259867,
-      "required_pages_per_second_upper_estimate": 0.35227319162943577
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27988.360192945398,
+      "required_pages_per_second_upper_estimate": 0.3475217625804959
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.29949027763285585,
-    "producer_p95_seconds": 0.4448636390006868,
-    "gpu_update_seconds": 0.9745153371123496,
-    "producer_batches_per_second": 3.3390065544160894,
-    "gpu_batches_per_second": 1.0261511152437741,
-    "producer_headroom": 3.25391309799714,
+    "producer_batch_seconds": 0.13427720356241934,
+    "producer_p95_seconds": 0.4788676940006553,
+    "gpu_update_seconds": 0.9878392234986604,
+    "producer_batches_per_second": 7.447280502346369,
+    "gpu_batches_per_second": 1.0123104815156756,
+    "producer_headroom": 7.356715788614551,
     "sustainable": true,
-    "buffer_seconds": 60.419950900965674,
+    "buffer_seconds": 60.25819263341828,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 457607.9721199526,
-    "upload_bytes_per_second": 5886509.698262579,
+    "generation_bytes_per_second": 460186.63650968054,
+    "upload_bytes_per_second": 5807437.331423879,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.19201824854788,
+    "estimated_full_shards_per_hour": 98.74533959834874,
     "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": 1.295291658003407,
-      "bytes": 11877354,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.0650672470001155,
       "bytes": 2271023,
@@ -562,6 +556,12 @@
       "bytes": 7313193,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1349576159991557,
+      "bytes": 5303247,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -427,3 +427,4 @@
 {"step": 157700, "loss": 0.5978798642754555, "lr": 0.0003, "grad_norm": 0.13798005187191756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23080472965, "tokens_per_second": 148117.4987348286, "data_wait_s": 1.2830998457502574e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 157800, "loss": 0.6031405199319124, "lr": 0.0003, "grad_norm": 0.13608469932209571, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23094887446, "tokens_per_second": 146077.91270982343, "data_wait_s": 7.855996955186129e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 157900, "loss": 0.6430932730436325, "lr": 0.0003, "grad_norm": 0.16743698220291053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23109366651, "tokens_per_second": 146229.2040780792, "data_wait_s": 8.094000804703683e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 158000, "loss": 0.5852047698572278, "lr": 0.0003, "grad_norm": 0.14220591175046893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23124097242, "tokens_per_second": 149570.96284369592, "data_wait_s": 1.2089003575965762e-05, "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:b1111f1cfc31e68b1f2f4177ddc916bdb0187023f9be250753574ade45cb01a1
-size 915051
+oid sha256:d6de3c6d8df23f33b8c9d48a2d43edf7ec27649ee17d7880f7292ab0211fb074
+size 917169