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": 75906.1754913118,
-      "required_pages_per_second_upper_estimate": 0.00657935490914397
+      "required_token_positions_per_second": 75254.61592427621,
+      "required_pages_per_second_upper_estimate": 0.00652287937723595
     },
     {
       "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": 9488.271936413976,
-      "required_pages_per_second_upper_estimate": 0.0037850135377429294
+      "required_token_positions_per_second": 9406.826990534526,
+      "required_pages_per_second_upper_estimate": 0.003752523931121161
     },
     {
       "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": 18976.54387282795,
-      "required_pages_per_second_upper_estimate": 0.0043898894492132735
+      "required_token_positions_per_second": 18813.653981069052,
+      "required_pages_per_second_upper_estimate": 0.004352207739519035
     },
     {
       "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": 28464.815809241925,
-      "required_pages_per_second_upper_estimate": 0.0017333579394361442
+      "required_token_positions_per_second": 28220.480971603574,
+      "required_pages_per_second_upper_estimate": 0.0017184792297497893
     },
     {
       "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": 28464.815809241925,
-      "required_pages_per_second_upper_estimate": 0.0021913003166487495
+      "required_token_positions_per_second": 28220.480971603574,
+      "required_pages_per_second_upper_estimate": 0.0021724907444852296
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28464.815809241925,
-      "required_pages_per_second_upper_estimate": 0.35343774674052825
+      "required_token_positions_per_second": 28220.480971603574,
+      "required_pages_per_second_upper_estimate": 0.3504039257931581
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.27542949726552024,
-    "producer_p95_seconds": 0.37525763400117285,
-    "gpu_update_seconds": 0.9713043704650735,
-    "producer_batches_per_second": 3.6306931898291834,
-    "gpu_batches_per_second": 1.0295433958782525,
-    "producer_headroom": 3.5265081630988644,
+    "producer_batch_seconds": 0.13274308461726037,
+    "producer_p95_seconds": 0.3414745040026901,
+    "gpu_update_seconds": 0.9797139895602904,
+    "producer_batches_per_second": 7.533349122353991,
+    "gpu_batches_per_second": 1.0207060536604302,
+    "producer_headroom": 7.380527523411941,
     "sustainable": true,
-    "buffer_seconds": 60.220870968834554,
+    "buffer_seconds": 60.742267352738004,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 473006.1320683427,
-    "upload_bytes_per_second": 4975434.77988957,
+    "generation_bytes_per_second": 472959.45254863735,
+    "upload_bytes_per_second": 5057322.454083147,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.49610492265425,
+    "estimated_full_shards_per_hour": 101.48608858436909,
     "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": 1.5228881869988982,
-      "bytes": 5895199,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0030821439977444,
       "bytes": 1174580,
@@ -562,6 +556,12 @@
       "bytes": 7374926,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.311550709000585,
+      "bytes": 11722471,
+      "pages": 42,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -290,3 +290,4 @@
 {"step": 144000, "loss": 0.6275187488645315, "lr": 0.0003, "grad_norm": 0.14284842503990988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21074305231, "tokens_per_second": 148443.2199237849, "data_wait_s": 1.4444998669205233e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 144100, "loss": 0.6331565584987402, "lr": 0.0003, "grad_norm": 0.16266841285290523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21088871629, "tokens_per_second": 148523.18870412576, "data_wait_s": 1.2602999049704522e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 144200, "loss": 0.6194612067192793, "lr": 0.0003, "grad_norm": 0.14854852839518176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21103405036, "tokens_per_second": 147933.94403916496, "data_wait_s": 1.1969001207035035e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 144300, "loss": 0.5910980794578791, "lr": 0.0003, "grad_norm": 0.15748691865434894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21118019654, "tokens_per_second": 149716.23294407126, "data_wait_s": 9.985000360757113e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8d57861fcc8945fd7c7a28d6e092ec166361a89d7bb016ca1f14f6fa692da62e
-size 622536
+oid sha256:df8a3cf4c54c1697f136c92fc83901d6afb9c173549461335d4cc19abd5de6d3
+size 624654