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": 75111.75112380728,
-      "required_pages_per_second_upper_estimate": 0.006510496218418829
+      "required_token_positions_per_second": 74648.82498241174,
+      "required_pages_per_second_upper_estimate": 0.006470370953758243
     },
     {
       "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": 9388.96889047591,
-      "required_pages_per_second_upper_estimate": 0.0037454000680053894
+      "required_token_positions_per_second": 9331.103122801467,
+      "required_pages_per_second_upper_estimate": 0.0037223165481097284
     },
     {
       "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": 18777.93778095182,
-      "required_pages_per_second_upper_estimate": 0.004343945425205566
+      "required_token_positions_per_second": 18662.206245602934,
+      "required_pages_per_second_upper_estimate": 0.0043171729713080585
     },
     {
       "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": 28166.906671427725,
-      "required_pages_per_second_upper_estimate": 0.0017152168359517099
+      "required_token_positions_per_second": 27993.309368404403,
+      "required_pages_per_second_upper_estimate": 0.0017046456709922486
     },
     {
       "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": 28166.906671427725,
-      "required_pages_per_second_upper_estimate": 0.0021683664465545383
+      "required_token_positions_per_second": 27993.309368404403,
+      "required_pages_per_second_upper_estimate": 0.0021550024456197117
     },
     {
       "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": 28166.906671427725,
-      "required_pages_per_second_upper_estimate": 0.3497387122866226
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27993.309368404403,
+      "required_pages_per_second_upper_estimate": 0.347583214775872
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.491650964882524,
-    "producer_p95_seconds": 0.3360536649997812,
-    "gpu_update_seconds": 0.9815774349139268,
-    "producer_batches_per_second": 2.03396326139407,
-    "gpu_batches_per_second": 1.0187683257894866,
-    "producer_headroom": 1.9964924408283562,
+    "producer_batch_seconds": 0.12434581860190974,
+    "producer_p95_seconds": 0.37025129300309345,
+    "gpu_update_seconds": 0.9876645749932608,
+    "producer_batches_per_second": 8.042087874313465,
+    "gpu_batches_per_second": 1.012489488151201,
+    "producer_headroom": 7.9428853024422645,
     "sustainable": true,
-    "buffer_seconds": 60.85780096466346,
+    "buffer_seconds": 60.24753907458891,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 264693.331985013,
-    "upload_bytes_per_second": 5319703.258108587,
+    "generation_bytes_per_second": 266001.22516544705,
+    "upload_bytes_per_second": 5351324.061289914,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.797027298572466,
+    "estimated_full_shards_per_hour": 57.0776707288986,
     "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": 0.9136627450025117,
-      "bytes": 6235523,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.3885049839991552,
       "bytes": 8289862,
@@ -562,6 +556,12 @@
       "bytes": 6950847,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3214142249998986,
+      "bytes": 10934190,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -339,3 +339,4 @@
 {"step": 388900, "loss": 0.5778171997517347, "lr": 0.0003, "grad_norm": 0.16686476276738277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56924420895, "tokens_per_second": 146948.23268215667, "data_wait_s": 1.3607001164928079e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 389000, "loss": 0.5543361138552427, "lr": 0.0003, "grad_norm": 0.13831569021399448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56939050728, "tokens_per_second": 126040.32964440901, "data_wait_s": 1.6482998034916818e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 389100, "loss": 0.5842744335532188, "lr": 0.0003, "grad_norm": 0.15831724049035797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56953725027, "tokens_per_second": 148946.78288122185, "data_wait_s": 1.166199945146218e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 389200, "loss": 0.5844362117350101, "lr": 0.0003, "grad_norm": 0.15270720085827424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56968332127, "tokens_per_second": 149604.25949259973, "data_wait_s": 1.1983996955677867e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fc7bcd1afbb1537574f271c487d4ddf91a4a36599a75ca2633914b50645afec0
-size 727101
+oid sha256:3e3022eaba14919715ef6b7d7d1c469729ba7bd11a3e0b26198348458dacc459
+size 729219