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": 74569.75015202582,
-      "required_pages_per_second_upper_estimate": 0.006463516947873717
+      "required_token_positions_per_second": 75735.892876488,
+      "required_pages_per_second_upper_estimate": 0.006564595243668384
     },
     {
       "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": 9321.218769003228,
-      "required_pages_per_second_upper_estimate": 0.0037183735315953516
+      "required_token_positions_per_second": 9466.986609561,
+      "required_pages_per_second_upper_estimate": 0.0037765225026172807
     },
     {
       "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": 18642.437538006456,
-      "required_pages_per_second_upper_estimate": 0.004312599828722984
+      "required_token_positions_per_second": 18933.973219122,
+      "required_pages_per_second_upper_estimate": 0.004380041477696318
     },
     {
       "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": 27963.65630700968,
-      "required_pages_per_second_upper_estimate": 0.0017028399551308995
+      "required_token_positions_per_second": 28400.959828682997,
+      "required_pages_per_second_upper_estimate": 0.0017294694452465432
     },
     {
       "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": 27963.65630700968,
-      "required_pages_per_second_upper_estimate": 0.0021527196708686185
+      "required_token_positions_per_second": 28400.959828682997,
+      "required_pages_per_second_upper_estimate": 0.0021863845065006556
     },
     {
       "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": 27963.65630700968,
-      "required_pages_per_second_upper_estimate": 0.34721502299576196
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28400.959828682997,
+      "required_pages_per_second_upper_estimate": 0.3526448691742056
     }
   ],
   "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": 0.074832562304735,
-    "producer_p95_seconds": 0.19214319799903024,
-    "gpu_update_seconds": 0.988711908645131,
-    "producer_batches_per_second": 13.363166637643321,
-    "gpu_batches_per_second": 1.011416967122746,
-    "producer_headroom": 13.212321991847265,
+    "producer_batch_seconds": 0.22433299817184604,
+    "producer_p95_seconds": 0.19128827699933026,
+    "gpu_update_seconds": 0.9734882259886668,
+    "producer_batches_per_second": 4.457658963011625,
+    "gpu_batches_per_second": 1.0272337900999347,
+    "producer_headroom": 4.339478515964667,
     "sustainable": true,
-    "buffer_seconds": 60.31142642735299,
+    "buffer_seconds": 60.35627001129734,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299588.0660923766,
-    "upload_bytes_per_second": 4058653.93467418,
+    "generation_bytes_per_second": 297349.5046879875,
+    "upload_bytes_per_second": 4063181.2369461562,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.28462493017648,
+    "estimated_full_shards_per_hour": 63.80428176383704,
     "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": 1.8453115359998264,
-      "bytes": 9879354,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.0517817119998654,
       "bytes": 7714712,
@@ -562,6 +556,12 @@
       "bytes": 7375397,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.9071520359993883,
+      "bytes": 6539230,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -98,3 +98,4 @@
 {"step": 324800, "loss": 0.5679438281804323, "lr": 0.0003, "grad_norm": 0.1692994539482135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47541583873, "tokens_per_second": 149203.58725814408, "data_wait_s": 1.4347999240271747e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 324900, "loss": 0.5770327933132648, "lr": 0.0003, "grad_norm": 0.15126750884586188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47556306635, "tokens_per_second": 150053.71359794465, "data_wait_s": 1.119599983212538e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 325000, "loss": 0.6160995494574308, "lr": 0.0003, "grad_norm": 0.14224377468636584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47571096641, "tokens_per_second": 151367.18161095746, "data_wait_s": 1.3118000424583443e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 325100, "loss": 0.6068964377045631, "lr": 0.0003, "grad_norm": 0.1606168216520234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47585915270, "tokens_per_second": 134267.51756507542, "data_wait_s": 1.288900057261344e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:66fed8601c3d7162da3ca9f1d95d6bfd6e119bc80708b31fc3be0f3af01ffbcd
-size 213531
+oid sha256:0013c399d50bea7a9343bdd9cf00019a3e79a8b6eadaf1a909369cb42303c093
+size 215649