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": 73955.25385548656,
-      "required_pages_per_second_upper_estimate": 0.006410253966316319
+      "required_token_positions_per_second": 75624.54132702427,
+      "required_pages_per_second_upper_estimate": 0.00655494357357879
     },
     {
       "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": 9244.40673193582,
-      "required_pages_per_second_upper_estimate": 0.003687732061566866
+      "required_token_positions_per_second": 9453.067665878034,
+      "required_pages_per_second_upper_estimate": 0.0037709700278753923
     },
     {
       "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": 18488.81346387164,
-      "required_pages_per_second_upper_estimate": 0.004277061602863256
+      "required_token_positions_per_second": 18906.135331756068,
+      "required_pages_per_second_upper_estimate": 0.004373601672384294
     },
     {
       "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": 27733.22019580746,
-      "required_pages_per_second_upper_estimate": 0.0016888076049635227
+      "required_token_positions_per_second": 28359.2029976341,
+      "required_pages_per_second_upper_estimate": 0.0017269266733168265
     },
     {
       "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": 27733.22019580746,
-      "required_pages_per_second_upper_estimate": 0.0021349800611402884
+      "required_token_positions_per_second": 28359.2029976341,
+      "required_pages_per_second_upper_estimate": 0.002183169950056205
     },
     {
       "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": 27733.22019580746,
-      "required_pages_per_second_upper_estimate": 0.3443537777146834
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28359.2029976341,
+      "required_pages_per_second_upper_estimate": 0.3521263890837019
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4183352053047713,
-    "producer_p95_seconds": 0.22930077799901483,
-    "gpu_update_seconds": 0.9969271438655241,
-    "producer_batches_per_second": 2.3904275502499637,
-    "gpu_batches_per_second": 1.0030823276840082,
-    "producer_headroom": 2.3830821102881576,
+    "producer_batch_seconds": 0.16101675750755362,
+    "producer_p95_seconds": 0.19550621899907128,
+    "gpu_update_seconds": 0.9749216154736988,
+    "producer_batches_per_second": 6.210533707667588,
+    "gpu_batches_per_second": 1.025723488051002,
+    "producer_headroom": 6.054783555233146,
     "sustainable": true,
-    "buffer_seconds": 60.812555775796966,
+    "buffer_seconds": 60.44514015936933,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 455926.627803381,
-    "upload_bytes_per_second": 5826980.028211579,
+    "generation_bytes_per_second": 434612.9482209978,
+    "upload_bytes_per_second": 5995992.47168965,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.83124089790414,
+    "estimated_full_shards_per_hour": 93.25782141659212,
     "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": [
@@ -376,6 +376,12 @@
       "bytes": 6307262,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1597189560015977,
+      "bytes": 13015122,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -87,3 +87,4 @@
 {"step": 93700, "loss": 0.620792331174016, "lr": 0.0003, "grad_norm": 0.13391458323998845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13712286885, "tokens_per_second": 150802.41198038854, "data_wait_s": 1.0600000678095967e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 93800, "loss": 0.6461169254034758, "lr": 0.0003, "grad_norm": 0.14760715023430984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13726921760, "tokens_per_second": 148834.92666011828, "data_wait_s": 1.0416999430162832e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 93900, "loss": 0.6738225184381008, "lr": 0.0003, "grad_norm": 0.1397822452711128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13741760457, "tokens_per_second": 151721.3559334485, "data_wait_s": 9.974999557016417e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 94000, "loss": 0.6277541741728783, "lr": 0.0003, "grad_norm": 0.14381173255763774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13756436345, "tokens_per_second": 149511.49512423552, "data_wait_s": 1.3134998880559579e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a030d2d1d0c2fe722891c00d726b6bb8850eee15e5177712f2414973035d921b
-size 189450
+oid sha256:17a1da80be775120e75d0e4b453ad5a14615d8914a781aaa59fa4f868a1b375a
+size 191568