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": 74483.50199021689,
-      "required_pages_per_second_upper_estimate": 0.0064560411758020865
+      "required_token_positions_per_second": 75405.28619997897,
+      "required_pages_per_second_upper_estimate": 0.006535939094863543
     },
     {
       "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": 9310.43774877711,
-      "required_pages_per_second_upper_estimate": 0.003714072821436537
+      "required_token_positions_per_second": 9425.660774997372,
+      "required_pages_per_second_upper_estimate": 0.0037600370093335616
     },
     {
       "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": 18620.87549755422,
-      "required_pages_per_second_upper_estimate": 0.004307611830143311
+      "required_token_positions_per_second": 18851.321549994744,
+      "required_pages_per_second_upper_estimate": 0.0043609214686634225
     },
     {
       "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": 27931.31324633133,
-      "required_pages_per_second_upper_estimate": 0.001700870432426513
+      "required_token_positions_per_second": 28276.982324992114,
+      "required_pages_per_second_upper_estimate": 0.0017219198657314602
     },
     {
       "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": 27931.31324633133,
-      "required_pages_per_second_upper_estimate": 0.0021502298125262768
+      "required_token_positions_per_second": 28276.982324992114,
+      "required_pages_per_second_upper_estimate": 0.002176840375074836
     },
     {
       "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": 27931.31324633133,
-      "required_pages_per_second_upper_estimate": 0.3468134304273977
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28276.982324992114,
+      "required_pages_per_second_upper_estimate": 0.35110548350437826
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.41722250333606326,
-    "producer_p95_seconds": 0.3993859699985478,
-    "gpu_update_seconds": 0.9898567874760223,
-    "producer_batches_per_second": 2.3968026460800047,
-    "gpu_batches_per_second": 1.0102471515600162,
-    "producer_headroom": 2.3724913674627834,
+    "producer_batch_seconds": 0.43244350320338754,
+    "producer_p95_seconds": 0.39517517799686175,
+    "gpu_update_seconds": 0.9777563844061182,
+    "producer_batches_per_second": 2.3124407988381277,
+    "gpu_batches_per_second": 1.022749650064819,
+    "producer_headroom": 2.261003754625164,
     "sustainable": true,
-    "buffer_seconds": 60.381264036037365,
+    "buffer_seconds": 60.62089583317933,
     "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": 360193.1821702826,
-    "upload_bytes_per_second": 5460733.277335636,
+    "generation_bytes_per_second": 326832.945231867,
+    "upload_bytes_per_second": 5519490.773991887,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.28907202559813,
+    "estimated_full_shards_per_hour": 70.13074176518448,
     "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": 1.1466966719999618,
-      "bytes": 6019789,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0688578599983884,
       "bytes": 4660632,
@@ -562,6 +556,12 @@
       "bytes": 6015320,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.3533267960010562,
+      "bytes": 11703451,
+      "pages": 42,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -343,3 +343,4 @@
 {"step": 149300, "loss": 0.6023747101426125, "lr": 0.0003, "grad_norm": 0.1355562907851094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21849749986, "tokens_per_second": 151387.0455247404, "data_wait_s": 1.2168005923740566e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 149400, "loss": 0.6421392858028412, "lr": 0.0003, "grad_norm": 0.1559292048793093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21864381975, "tokens_per_second": 148578.63435140913, "data_wait_s": 1.1220996384508908e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 149500, "loss": 0.6123340763151646, "lr": 0.0003, "grad_norm": 0.1493292369982806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21878873365, "tokens_per_second": 142787.79323841928, "data_wait_s": 9.082999895326793e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 149600, "loss": 0.5529136750847101, "lr": 0.0003, "grad_norm": 0.13856035853496387, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21893515055, "tokens_per_second": 149580.10658214678, "data_wait_s": 1.0572002793196589e-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 sha256:0aa935e215d2a1c5dc53953ecd3dc2e2d4093cf0805cc57cdc2409e5fab76b80
-size 735573
+oid sha256:48b5b8149632cf7e528d7f10cd33715c446da5352d6b9154fef3f7cb47a6ed0b
+size 737691