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": 76141.73507905054,
-      "required_pages_per_second_upper_estimate": 0.006599772617188847
+      "required_token_positions_per_second": 75737.32730859044,
+      "required_pages_per_second_upper_estimate": 0.006564719576607488
     },
     {
       "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": 9517.716884881318,
-      "required_pages_per_second_upper_estimate": 0.0037967595679277637
+      "required_token_positions_per_second": 9467.165913573805,
+      "required_pages_per_second_upper_estimate": 0.003776594029668823
     },
     {
       "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": 19035.433769762636,
-      "required_pages_per_second_upper_estimate": 0.004403512590442325
+      "required_token_positions_per_second": 18934.33182714761,
+      "required_pages_per_second_upper_estimate": 0.004380124435351757
     },
     {
       "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": 28553.150654643956,
-      "required_pages_per_second_upper_estimate": 0.0017387370680640119
+      "required_token_positions_per_second": 28401.497740721414,
+      "required_pages_per_second_upper_estimate": 0.0017295022012674764
     },
     {
       "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": 28553.150654643956,
-      "required_pages_per_second_upper_estimate": 0.0021981005775743
+      "required_token_positions_per_second": 28401.497740721414,
+      "required_pages_per_second_upper_estimate": 0.002186425916458409
     },
     {
       "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": 28553.150654643956,
-      "required_pages_per_second_upper_estimate": 0.35453456988271176
+      "required_token_positions_per_second": 28401.497740721414,
+      "required_pages_per_second_upper_estimate": 0.35265154824144695
     }
   ],
   "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": 0.09405767789826314,
-    "producer_p95_seconds": 0.25813985700006015,
-    "gpu_update_seconds": 0.9682994473852664,
-    "producer_batches_per_second": 10.63177427239532,
-    "gpu_batches_per_second": 1.0327383772657681,
-    "producer_headroom": 10.294741152685281,
+    "producer_batch_seconds": 0.08668604147669612,
+    "producer_p95_seconds": 0.2792946830013534,
+    "gpu_update_seconds": 0.9734697885442476,
+    "producer_batches_per_second": 11.535882628448674,
+    "gpu_batches_per_second": 1.0272532458304908,
+    "producer_headroom": 11.22983322298719,
     "sustainable": true,
-    "buffer_seconds": 60.03456573788652,
+    "buffer_seconds": 60.35512688974335,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387091.66837291163,
-    "upload_bytes_per_second": 6303212.895463713,
+    "generation_bytes_per_second": 430058.8266814404,
+    "upload_bytes_per_second": 6326863.173314761,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.06086099997054,
+    "estimated_full_shards_per_hour": 92.28061294872674,
     "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": 1.1437757880012214,
-      "bytes": 6056576,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.102731920998849,
       "bytes": 2598092,
@@ -562,6 +556,12 @@
       "bytes": 2306768,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.8912073430001328,
+      "bytes": 6007286,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -203,3 +203,4 @@
 {"step": 60300, "loss": 0.6200925186276436, "lr": 0.0003, "grad_norm": 0.16018672564577316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8825286552, "tokens_per_second": 148037.3266823668, "data_wait_s": 1.9115999748464674e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 60400, "loss": 0.6006607227027416, "lr": 0.0003, "grad_norm": 0.1277029163917345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8839879911, "tokens_per_second": 149213.90652760616, "data_wait_s": 9.121002221945673e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 60500, "loss": 0.6174721997231245, "lr": 0.0003, "grad_norm": 0.14301184221771793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8854581663, "tokens_per_second": 149821.39548354232, "data_wait_s": 1.1247997463215142e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 60600, "loss": 0.596485260874033, "lr": 0.0003, "grad_norm": 0.12365103573619238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8869240999, "tokens_per_second": 148779.7304257529, "data_wait_s": 1.3098000636091456e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2256bb48748555374fbaa8095ef1bed0511959c253b294a9f683677edc076be2
-size 437487
+oid sha256:c59663663c12b8b403862fc61058e2f6146569e82d13acd1b9943269b8b969b4
+size 439605