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": 75105.55591063014,
-      "required_pages_per_second_upper_estimate": 0.006509959233042259
+      "required_token_positions_per_second": 74227.79907606945,
+      "required_pages_per_second_upper_estimate": 0.006433877495276898
     },
     {
       "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": 9388.194488828767,
-      "required_pages_per_second_upper_estimate": 0.0037450911476100074
+      "required_token_positions_per_second": 9278.474884508681,
+      "required_pages_per_second_upper_estimate": 0.0037013223569924533
     },
     {
       "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": 18776.388977657534,
-      "required_pages_per_second_upper_estimate": 0.004343587136821441
+      "required_token_positions_per_second": 18556.949769017363,
+      "required_pages_per_second_upper_estimate": 0.004292823737900705
     },
     {
       "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": 28164.583466486303,
-      "required_pages_per_second_upper_estimate": 0.0017150753649596915
+      "required_token_positions_per_second": 27835.424653526046,
+      "required_pages_per_second_upper_estimate": 0.0016950313202132368
     },
     {
       "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": 28164.583466486303,
-      "required_pages_per_second_upper_estimate": 0.0021681875998071044
+      "required_token_positions_per_second": 27835.424653526046,
+      "required_pages_per_second_upper_estimate": 0.002142848043215512
     },
     {
       "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": 28164.583466486303,
-      "required_pages_per_second_upper_estimate": 0.3497098658565169
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27835.424653526046,
+      "required_pages_per_second_upper_estimate": 0.3456228150232321
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.3911090248360125,
-    "producer_p95_seconds": 0.22916250600064814,
-    "gpu_update_seconds": 0.9816584020459241,
-    "producer_batches_per_second": 2.5568318205372234,
-    "gpu_batches_per_second": 1.0186842978329826,
-    "producer_headroom": 2.509935439248742,
+    "producer_batch_seconds": 0.07696638953127888,
+    "producer_p95_seconds": 0.1970750059990678,
+    "gpu_update_seconds": 0.9932666860355477,
+    "producer_batches_per_second": 12.992684288427007,
+    "gpu_batches_per_second": 1.0067789588225566,
+    "producer_headroom": 12.905200465872023,
     "sustainable": true,
-    "buffer_seconds": 60.8628209268473,
+    "buffer_seconds": 60.58926784816841,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297233.6651835419,
-    "upload_bytes_per_second": 3917580.147371471,
+    "generation_bytes_per_second": 302466.97248374246,
+    "upload_bytes_per_second": 4021058.617092082,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.77942530278866,
+    "estimated_full_shards_per_hour": 64.90237122425275,
     "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": 3.6869929439999396,
-      "bytes": 11547756,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 0.8728801329998532,
       "bytes": 1038671,
@@ -562,6 +556,12 @@
       "bytes": 975468,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.1824829600009252,
+      "bytes": 12495213,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -94,3 +94,4 @@
 {"step": 324400, "loss": 0.5966446306556463, "lr": 0.0003, "grad_norm": 0.19220411014517647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47482978852, "tokens_per_second": 147404.66821660692, "data_wait_s": 1.2679000064963475e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 324500, "loss": 0.566635305993259, "lr": 0.0003, "grad_norm": 0.1616167549320805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47497585017, "tokens_per_second": 148109.14140868926, "data_wait_s": 1.6760001017246395e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 324600, "loss": 0.6210855878889561, "lr": 0.0003, "grad_norm": 0.1430476559778771, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47512305924, "tokens_per_second": 149788.89772493474, "data_wait_s": 6.219999704626389e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 324700, "loss": 0.5599087551236153, "lr": 0.0003, "grad_norm": 0.1449137258835772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47526927889, "tokens_per_second": 148547.9794566315, "data_wait_s": 1.4236999049899168e-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 sha256:9019d57374b89911f1547303fe68d405515de234a27241cdbfca45034330bd2c
-size 204276
+oid sha256:7f91c82834cf9ac408eb4cc1b070d9751eb686409f7a70de5f95d3b898c5b5be
+size 206394