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": 77171.99133999672,
-      "required_pages_per_second_upper_estimate": 0.005338937424591044
+      "required_token_positions_per_second": 76798.98710084285,
+      "required_pages_per_second_upper_estimate": 0.005313132125837303
     },
     {
       "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": 9646.49891749959,
-      "required_pages_per_second_upper_estimate": 0.00403878765495297
+      "required_token_positions_per_second": 9599.873387605356,
+      "required_pages_per_second_upper_estimate": 0.004019266519238036
     },
     {
       "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": 19292.99783499918,
-      "required_pages_per_second_upper_estimate": 0.004018222658462737
+      "required_token_positions_per_second": 19199.74677521071,
+      "required_pages_per_second_upper_estimate": 0.003998800921904619
     },
     {
       "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": 28939.49675249877,
-      "required_pages_per_second_upper_estimate": 0.0017348786422593315
+      "required_token_positions_per_second": 28799.620162816063,
+      "required_pages_per_second_upper_estimate": 0.001726493254287039
     },
     {
       "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": 28939.49675249877,
-      "required_pages_per_second_upper_estimate": 0.0022499185811131342
+      "required_token_positions_per_second": 28799.620162816063,
+      "required_pages_per_second_upper_estimate": 0.0022390437915174
     },
     {
       "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": 28939.49675249877,
-      "required_pages_per_second_upper_estimate": 0.3712047915303648
+      "required_token_positions_per_second": 28799.620162816063,
+      "required_pages_per_second_upper_estimate": 0.3694106048256957
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.45373727575034195,
-    "producer_p95_seconds": 0.28992689100050484,
-    "gpu_update_seconds": 0.9553725220744463,
-    "producer_batches_per_second": 2.2039185525286795,
-    "gpu_batches_per_second": 1.0467121221245215,
-    "producer_headroom": 2.1055632259759878,
+    "producer_batch_seconds": 0.09263921337509373,
+    "producer_p95_seconds": 0.25878172000011546,
+    "gpu_update_seconds": 0.9600126614064531,
+    "producer_batches_per_second": 10.794564888531884,
+    "gpu_batches_per_second": 1.041652928342595,
+    "producer_headroom": 10.362918967364147,
     "sustainable": true,
-    "buffer_seconds": 60.18846889069012,
+    "buffer_seconds": 60.48079766860655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362935.1925405347,
-    "upload_bytes_per_second": 5710459.827395127,
+    "generation_bytes_per_second": 377079.4695993611,
+    "upload_bytes_per_second": 5691200.689800616,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.87744362031967,
+    "estimated_full_shards_per_hour": 80.91247621522547,
     "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": 1.1479466919990955,
-      "bytes": 10073767,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.986695225998119,
       "bytes": 1193548,
@@ -562,6 +556,12 @@
       "bytes": 10185249,
       "pages": 22,
       "shards": 3
+    },
+    {
+      "seconds": 1.1744725909993576,
+      "bytes": 8758656,
+      "pages": 47,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -231,3 +231,4 @@
 {"step": 508100, "loss": 0.6056517362594604, "lr": 0.0003, "grad_norm": 0.16544024449387015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74368753894, "tokens_per_second": 153186.17762033496, "data_wait_s": 8.586001058574766e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 508200, "loss": 0.5989865064620972, "lr": 0.0003, "grad_norm": 0.1722002195965079, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74383504709, "tokens_per_second": 153661.78541964802, "data_wait_s": 8.799001079751179e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 508300, "loss": 0.590277373790741, "lr": 0.0003, "grad_norm": 0.15147490358940513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74398270775, "tokens_per_second": 153651.1261466149, "data_wait_s": 1.5434998204000294e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 508400, "loss": 0.5920135974884033, "lr": 0.0003, "grad_norm": 0.149531166071458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74412898822, "tokens_per_second": 152571.57669540404, "data_wait_s": 8.951999916462228e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:baf10f709322777e163abe99a26911f91b61d265b95b8ddb17189a41aa950d06
-size 496791
+oid sha256:632ed0cf44fe9808c2e77172c6b1780b96e8d77fe8f90985ac64b400018cbd1f
+size 498909