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": 77182.61218004843,
-      "required_pages_per_second_upper_estimate": 0.005339672198949565
+      "required_token_positions_per_second": 77208.6757034937,
+      "required_pages_per_second_upper_estimate": 0.005341475333977216
     },
     {
       "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": 9647.826522506053,
-      "required_pages_per_second_upper_estimate": 0.004039343495445631
+      "required_token_positions_per_second": 9651.084462936713,
+      "required_pages_per_second_upper_estimate": 0.004040707527070415
     },
     {
       "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": 19295.653045012106,
-      "required_pages_per_second_upper_estimate": 0.004018775668685832
+      "required_token_positions_per_second": 19302.168925873426,
+      "required_pages_per_second_upper_estimate": 0.004020132754833912
     },
     {
       "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": 28943.47956751816,
-      "required_pages_per_second_upper_estimate": 0.0017351174059383374
+      "required_token_positions_per_second": 28953.253388810142,
+      "required_pages_per_second_upper_estimate": 0.0017357033316010308
     },
     {
       "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": 28943.47956751816,
-      "required_pages_per_second_upper_estimate": 0.00225022822746234
+      "required_token_positions_per_second": 28953.253388810142,
+      "required_pages_per_second_upper_estimate": 0.002250988099077295
     },
     {
       "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": 28943.47956751816,
-      "required_pages_per_second_upper_estimate": 0.3712558788050199
+      "required_token_positions_per_second": 28953.253388810142,
+      "required_pages_per_second_upper_estimate": 0.3713812468902418
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10312117828910061,
-    "producer_p95_seconds": 0.2058258749984816,
-    "gpu_update_seconds": 0.9552410564702106,
-    "producer_batches_per_second": 9.697329070431064,
-    "gpu_batches_per_second": 1.046856176487202,
-    "producer_headroom": 9.263286866177854,
+    "producer_batch_seconds": 0.10196029207808976,
+    "producer_p95_seconds": 0.2100582890016085,
+    "gpu_update_seconds": 0.9549185933863102,
+    "producer_batches_per_second": 9.807739656474462,
+    "gpu_batches_per_second": 1.047209685648515,
+    "producer_headroom": 9.365592957059727,
     "sustainable": true,
-    "buffer_seconds": 60.18018655762327,
+    "buffer_seconds": 60.159871383337546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 423385.32330597093,
-    "upload_bytes_per_second": 5296435.742649299,
+    "generation_bytes_per_second": 442755.59976338595,
+    "upload_bytes_per_second": 5221791.541573747,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.84863447555873,
+    "estimated_full_shards_per_hour": 95.00504488636193,
     "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": 1.2959960469997895,
-      "bytes": 13446052,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.8787024550001661,
       "bytes": 837425,
@@ -562,6 +556,12 @@
       "bytes": 5791973,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.1227300080008717,
+      "bytes": 6451625,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -136,3 +136,4 @@
 {"step": 618000, "loss": 0.5506353974342346, "lr": 0.0001, "grad_norm": 0.1530333902383697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90457754751, "tokens_per_second": 152628.27553322975, "data_wait_s": 1.4923998605809174e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 618100, "loss": 0.5441434383392334, "lr": 0.0001, "grad_norm": 0.16083786768726563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90472583856, "tokens_per_second": 154745.39214966018, "data_wait_s": 1.2869999409304e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 618200, "loss": 0.5408501625061035, "lr": 0.0001, "grad_norm": 0.1663351494134956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90487221359, "tokens_per_second": 152275.61263524686, "data_wait_s": 1.2533999324659817e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 618300, "loss": 0.4924952983856201, "lr": 0.0001, "grad_norm": 0.15346313072702455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90501707757, "tokens_per_second": 151103.03315400836, "data_wait_s": 1.0628999007167295e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fecadd463568541250c2de1bcb14ee6494aa8e59bb260a901e9b26cf8ac66736
-size 294798
+oid sha256:5bc4cec9dcd5d9ee2c9733b9e6cd8c02db0d2ab8d847f409fb2ae7f48abee3f7
+size 296916