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": 74921.68402047131,
-      "required_pages_per_second_upper_estimate": 0.006494021683622337
+      "required_token_positions_per_second": 75288.38925077853,
+      "required_pages_per_second_upper_estimate": 0.006525806763579458
     },
     {
       "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": 9365.210502558914,
-      "required_pages_per_second_upper_estimate": 0.003735922491845745
+      "required_token_positions_per_second": 9411.048656347317,
+      "required_pages_per_second_upper_estimate": 0.0037542080167334117
     },
     {
       "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": 18730.421005117827,
-      "required_pages_per_second_upper_estimate": 0.004332953255383065
+      "required_token_positions_per_second": 18822.097312694634,
+      "required_pages_per_second_upper_estimate": 0.004354160955693052
     },
     {
       "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": 28095.631507676742,
-      "required_pages_per_second_upper_estimate": 0.0017108765524311586
+      "required_token_positions_per_second": 28233.145969041954,
+      "required_pages_per_second_upper_estimate": 0.0017192504616456803
     },
     {
       "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": 28095.631507676742,
-      "required_pages_per_second_upper_estimate": 0.002162879487146706
+      "required_token_positions_per_second": 28233.145969041954,
+      "required_pages_per_second_upper_estimate": 0.0021734657310469928
     },
     {
       "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": 28095.631507676742,
-      "required_pages_per_second_upper_estimate": 0.3488537132954635
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28233.145969041954,
+      "required_pages_per_second_upper_estimate": 0.350561182674323
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.12218737527356893,
-    "producer_p95_seconds": 0.3115486620008596,
-    "gpu_update_seconds": 0.9840675762153831,
-    "producer_batches_per_second": 8.184151576716255,
-    "gpu_batches_per_second": 1.0161903757116877,
-    "producer_headroom": 8.053758205478472,
+    "producer_batch_seconds": 0.06685185610959365,
+    "producer_p95_seconds": 0.23646433600151795,
+    "gpu_update_seconds": 0.9792745034618681,
+    "producer_batches_per_second": 14.958447800770843,
+    "gpu_batches_per_second": 1.021164133718242,
+    "producer_headroom": 14.64842654266014,
     "sustainable": true,
-    "buffer_seconds": 60.02812214913837,
+    "buffer_seconds": 60.71501921463582,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356376.0322782526,
-    "upload_bytes_per_second": 5734460.895811567,
+    "generation_bytes_per_second": 373694.2227260017,
+    "upload_bytes_per_second": 5728428.090125546,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.4700005174702,
+    "estimated_full_shards_per_hour": 80.18608104071653,
     "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": 0.8906719839997095,
-      "bytes": 2274847,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.014783542999794,
       "bytes": 6341777,
@@ -562,6 +556,12 @@
       "bytes": 6178099,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0223706500000844,
+      "bytes": 2585543,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -207,3 +207,4 @@
 {"step": 375700, "loss": 0.5926074273884296, "lr": 0.0003, "grad_norm": 0.15002919995749728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54991231745, "tokens_per_second": 149483.97750365784, "data_wait_s": 1.0679999832063913e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 375800, "loss": 0.5588666722178459, "lr": 0.0003, "grad_norm": 0.15751835521936036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55006008728, "tokens_per_second": 149650.65477589727, "data_wait_s": 2.2377000277629122e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 375900, "loss": 0.5704503655433655, "lr": 0.0003, "grad_norm": 0.13811188316819115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55020796021, "tokens_per_second": 151098.66428587525, "data_wait_s": 1.5047000488266349e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 376000, "loss": 0.6168783605098724, "lr": 0.0003, "grad_norm": 0.1514677382010011, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55035484909, "tokens_per_second": 149685.17553079486, "data_wait_s": 1.2767002772307023e-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 sha256:19016be3f53a37295b7f8a8c991356afa3fb4e9afdd7fcc5b415eff7e2c2df32
-size 445959
+oid sha256:5203cb49066a6ef02cd8025448a4ea850ee1ef83699a0aec5272337b6c694e58
+size 448077