Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77516.3302752865,
-      "required_pages_per_second_upper_estimate": 0.005362759591110798
+      "required_token_positions_per_second": 77232.1223263962,
+      "required_pages_per_second_upper_estimate": 0.005343097425753271
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9689.541284410812,
-      "required_pages_per_second_upper_estimate": 0.004056808595151868
+      "required_token_positions_per_second": 9654.015290799525,
+      "required_pages_per_second_upper_estimate": 0.00404193460349393
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19379.082568821625,
-      "required_pages_per_second_upper_estimate": 0.004036151838310856
+      "required_token_positions_per_second": 19308.03058159905,
+      "required_pages_per_second_upper_estimate": 0.0040213535831393026
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29068.62385323244,
-      "required_pages_per_second_upper_estimate": 0.0017426196147826483
+      "required_token_positions_per_second": 28962.045872398576,
+      "required_pages_per_second_upper_estimate": 0.0017362304275668123
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29068.62385323244,
-      "required_pages_per_second_upper_estimate": 0.002259957645225086
+      "required_token_positions_per_second": 28962.045872398576,
+      "required_pages_per_second_upper_estimate": 0.002251671676002248
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29068.62385323244,
-      "required_pages_per_second_upper_estimate": 0.3728610953326976
+      "required_token_positions_per_second": 28962.045872398576,
+      "required_pages_per_second_upper_estimate": 0.3714940274290809
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05249039916405174,
-    "producer_p95_seconds": 0.11996662100000322,
-    "gpu_update_seconds": 0.9511286168755297,
-    "producer_batches_per_second": 19.05110298122583,
-    "gpu_batches_per_second": 1.051382517839715,
-    "producer_headroom": 18.120049228486607,
+    "producer_batch_seconds": 0.05331649055468901,
+    "producer_p95_seconds": 0.1171844630000578,
+    "gpu_update_seconds": 0.9546286930768627,
+    "producer_batches_per_second": 18.755923159913483,
+    "gpu_batches_per_second": 1.0475277008246013,
+    "producer_headroom": 17.904942413598267,
     "sustainable": true,
-    "buffer_seconds": 60.8722314800339,
+    "buffer_seconds": 60.14160766384235,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 64,
+    "upload_pages_per_commit": 63,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 885700, "loss": 0.5637478828430176, "lr": 3e-05, "grad_norm": 0.3168497868655922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129641887771, "tokens_per_second": 155400.9375584298, "data_wait_s": 8.966999985204893e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 885800, "loss": 0.5134460926055908, "lr": 3e-05, "grad_norm": 0.3260002510281071, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129656413302, "tokens_per_second": 152843.74385623902, "data_wait_s": 7.922000122562167e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 885900, "loss": 0.6446624398231506, "lr": 3e-05, "grad_norm": 0.3511060332315161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129670955519, "tokens_per_second": 152595.54922396873, "data_wait_s": 9.253999905922683e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 886000, "loss": 0.547597348690033, "lr": 3e-05, "grad_norm": 0.3115406683974517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129685560151, "tokens_per_second": 153690.18540588112, "data_wait_s": 1.1384999879737734e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:811b1e79a13dae6efdea92c6019817b51071886a87c360316c7e9bc0f4f038d1
-size 19227
+oid sha256:c57a922ef7131de3a2f63c92411c813248611f55ea0be8972eb98b076aa3d40e
+size 21345