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": 74345.90759310707,
-      "required_pages_per_second_upper_estimate": 0.006444114842190418
+      "required_token_positions_per_second": 75453.23133152456,
+      "required_pages_per_second_upper_estimate": 0.006540094857350089
     },
     {
       "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": 9293.238449138384,
-      "required_pages_per_second_upper_estimate": 0.0037072117636582033
+      "required_token_positions_per_second": 9431.65391644057,
+      "required_pages_per_second_upper_estimate": 0.0037624277630607028
     },
     {
       "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": 18586.476898276767,
-      "required_pages_per_second_upper_estimate": 0.004299654319595142
+      "required_token_positions_per_second": 18863.30783288114,
+      "required_pages_per_second_upper_estimate": 0.004363694284257816
     },
     {
       "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": 27879.71534741515,
-      "required_pages_per_second_upper_estimate": 0.0016977283910957711
+      "required_token_positions_per_second": 28294.961749321705,
+      "required_pages_per_second_upper_estimate": 0.001723014718342383
     },
     {
       "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": 27879.71534741515,
-      "required_pages_per_second_upper_estimate": 0.002146257663435583
+      "required_token_positions_per_second": 28294.961749321705,
+      "required_pages_per_second_upper_estimate": 0.00217822448092996
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27879.71534741515,
-      "required_pages_per_second_upper_estimate": 0.3461727572099178
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28294.961749321705,
+      "required_pages_per_second_upper_estimate": 0.35132872778128943
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13613870142972928,
-    "producer_p95_seconds": 0.35325613299937686,
-    "gpu_update_seconds": 0.9916887477319013,
-    "producer_batches_per_second": 7.345449820646115,
-    "gpu_batches_per_second": 1.0083809081096335,
-    "producer_headroom": 7.284399934164065,
+    "producer_batch_seconds": 0.4469071633047008,
+    "producer_p95_seconds": 0.3654016690015851,
+    "gpu_update_seconds": 0.9771350901600984,
+    "producer_batches_per_second": 2.2376011890375564,
+    "gpu_batches_per_second": 1.0233999475304436,
+    "producer_headroom": 2.186438639592556,
     "sustainable": true,
-    "buffer_seconds": 60.49301361164598,
+    "buffer_seconds": 60.5823755899261,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341130.2130052205,
-    "upload_bytes_per_second": 4545911.852563372,
+    "generation_bytes_per_second": 304288.57047011727,
+    "upload_bytes_per_second": 4570996.796675718,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.19860260598622,
+    "estimated_full_shards_per_hour": 65.29324374749793,
     "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": 1.2225116730005539,
-      "bytes": 6709538,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.5987381729974004,
       "bytes": 11965797,
@@ -562,6 +556,12 @@
       "bytes": 1226801,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 0.896007103001466,
+      "bytes": 7569788,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -292,3 +292,4 @@
 {"step": 184200, "loss": 0.5851279124617577, "lr": 0.0003, "grad_norm": 0.1448492210517399, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26959426724, "tokens_per_second": 150033.83465250413, "data_wait_s": 1.2533997505670413e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 184300, "loss": 0.6379616372287273, "lr": 0.0003, "grad_norm": 0.14567102136135143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26974147930, "tokens_per_second": 150217.05002835635, "data_wait_s": 1.0604002454783767e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 184400, "loss": 0.6136236861348152, "lr": 0.0003, "grad_norm": 0.15183699397497416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26988803415, "tokens_per_second": 148993.5823434023, "data_wait_s": 1.0564999683992937e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 184500, "loss": 0.5798609592020512, "lr": 0.0003, "grad_norm": 0.1502639114966204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27003277715, "tokens_per_second": 147708.54063929812, "data_wait_s": 1.5797999367350712e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7deaf14a67ff11b9ecb2aacb17f6b4e09bd7997c9417c420bcc014d57d773264
-size 626772
+oid sha256:f5de8a92434c528493c70f415736b1175689926c2bcc142d8804d722bdfa33a5
+size 628890