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": 75571.23727573993,
-      "required_pages_per_second_upper_estimate": 0.0065503233135113465
+      "required_token_positions_per_second": 74516.28837657048,
+      "required_pages_per_second_upper_estimate": 0.006458883016674879
     },
     {
       "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": 9446.404659467491,
-      "required_pages_per_second_upper_estimate": 0.0037683120549974036
+      "required_token_positions_per_second": 9314.53604707131,
+      "required_pages_per_second_upper_estimate": 0.0037157076939011127
     },
     {
       "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": 18892.809318934982,
-      "required_pages_per_second_upper_estimate": 0.0043705189338479515
+      "required_token_positions_per_second": 18629.07209414262,
+      "required_pages_per_second_upper_estimate": 0.004309507968508863
     },
     {
       "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": 28339.213978402473,
-      "required_pages_per_second_upper_estimate": 0.001725709446919899
+      "required_token_positions_per_second": 27943.60814121393,
+      "required_pages_per_second_upper_estimate": 0.001701619127018535
     },
     {
       "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": 28339.213978402473,
-      "required_pages_per_second_upper_estimate": 0.002181631140022608
+      "required_token_positions_per_second": 27943.60814121393,
+      "required_pages_per_second_upper_estimate": 0.0021511763075687863
     },
     {
       "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": 28339.213978402473,
-      "required_pages_per_second_upper_estimate": 0.351878192363789
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27943.60814121393,
+      "required_pages_per_second_upper_estimate": 0.34696609187347344
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 688946,
-    "producer_batch_seconds": 0.09231783152344519,
-    "producer_p95_seconds": 0.2204372449996299,
-    "gpu_update_seconds": 0.9756092748751164,
-    "producer_batches_per_second": 10.83214351439828,
-    "gpu_batches_per_second": 1.0250005055845801,
-    "producer_headroom": 10.5679396794253,
+    "producer_batch_seconds": 0.361804677031202,
+    "producer_p95_seconds": 0.2962313389998599,
+    "gpu_update_seconds": 0.989421260857937,
+    "producer_batches_per_second": 2.7639222582900995,
+    "gpu_batches_per_second": 1.0106918453853417,
+    "producer_headroom": 2.734683445710707,
     "sustainable": true,
-    "buffer_seconds": 60.48777504225722,
+    "buffer_seconds": 60.354696912334155,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351155.1813792319,
+    "generation_bytes_per_second": 342781.14988659945,
     "upload_bytes_per_second": 5211224.213534921,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.34972744972914,
+    "estimated_full_shards_per_hour": 73.55285522888649,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -154,3 +154,4 @@
 {"step": 210400, "loss": 0.6081433296203613, "lr": 0.0003, "grad_norm": 0.14212834516230086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30794933317, "tokens_per_second": 149589.68693309152, "data_wait_s": 1.61259995365981e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 210500, "loss": 0.6219476591795683, "lr": 0.0003, "grad_norm": 0.14513659955616465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30809731460, "tokens_per_second": 150822.22675566532, "data_wait_s": 1.0117000783793628e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 210600, "loss": 0.6202474236488342, "lr": 0.0003, "grad_norm": 0.1715747130926135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30824467153, "tokens_per_second": 149190.10893735292, "data_wait_s": 1.6257999959634617e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 210700, "loss": 0.5392297208309174, "lr": 0.0003, "grad_norm": 0.14981244715378203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30839089878, "tokens_per_second": 149101.70096932698, "data_wait_s": 1.6451000192319043e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:21d1f49a787962ab771617e0cea4ff2a35fb731d64ac0f05872df163c90974f8
-size 332922
+oid sha256:7969f4cfd622f4b22dfa632f8a0c84fd619a82e58949615d30bd1f8a6e530e14
+size 335040