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": 76893.01410705768,
-      "required_pages_per_second_upper_estimate": 0.005319637132299958
+      "required_token_positions_per_second": 76944.96289502086,
+      "required_pages_per_second_upper_estimate": 0.005323231069989045
     },
     {
       "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": 9611.62676338221,
-      "required_pages_per_second_upper_estimate": 0.0040241874122374085
+      "required_token_positions_per_second": 9618.120361877607,
+      "required_pages_per_second_upper_estimate": 0.004026906146325675
     },
     {
       "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": 19223.25352676442,
-      "required_pages_per_second_upper_estimate": 0.00400369675833853
+      "required_token_positions_per_second": 19236.240723755214,
+      "required_pages_per_second_upper_estimate": 0.004006401648976296
     },
     {
       "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": 28834.88029014663,
-      "required_pages_per_second_upper_estimate": 0.0017286070450814094
+      "required_token_positions_per_second": 28854.361085632823,
+      "required_pages_per_second_upper_estimate": 0.0017297748890253543
     },
     {
       "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": 28834.88029014663,
-      "required_pages_per_second_upper_estimate": 0.0022417851113244423
+      "required_token_positions_per_second": 28854.361085632823,
+      "required_pages_per_second_upper_estimate": 0.00224329965748653
     },
     {
       "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": 28834.88029014663,
-      "required_pages_per_second_upper_estimate": 0.36986288387971716
+      "required_token_positions_per_second": 28854.361085632823,
+      "required_pages_per_second_upper_estimate": 0.37011276260736553
     }
   ],
   "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": 0.13424891145322704,
-    "producer_p95_seconds": 0.2763498470012564,
-    "gpu_update_seconds": 0.9588387301003567,
-    "producer_batches_per_second": 7.448849969620832,
-    "gpu_batches_per_second": 1.0429282512350488,
-    "producer_headroom": 7.142245845579319,
+    "producer_batch_seconds": 0.0957610159686908,
+    "producer_p95_seconds": 0.2836874060012633,
+    "gpu_update_seconds": 0.9581913776550923,
+    "producer_batches_per_second": 10.4426628089133,
+    "gpu_batches_per_second": 1.0436328517662334,
+    "producer_headroom": 10.00606946326023,
     "sustainable": true,
-    "buffer_seconds": 60.406839996322475,
+    "buffer_seconds": 60.36605679227082,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 402820.629700422,
-    "upload_bytes_per_second": 4931203.727040912,
+    "generation_bytes_per_second": 404259.93902852375,
+    "upload_bytes_per_second": 4987444.309384191,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.43592994937416,
+    "estimated_full_shards_per_hour": 86.74477222577843,
     "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": 1.0456256510005915,
-      "bytes": 862691,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 2.013755919000687,
       "bytes": 11740969,
@@ -562,6 +556,12 @@
       "bytes": 13390262,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 0.8803229269979056,
+      "bytes": 4831371,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -248,3 +248,4 @@
 {"step": 421900, "loss": 0.6058818697929382, "lr": 0.0003, "grad_norm": 0.1551188317121638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61754736801, "tokens_per_second": 152753.96167593866, "data_wait_s": 1.0063999070553109e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 422000, "loss": 0.582892656326294, "lr": 0.0003, "grad_norm": 0.1720947304972548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61769492791, "tokens_per_second": 152094.79720752212, "data_wait_s": 9.811999916564673e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 422100, "loss": 0.5418153405189514, "lr": 0.0003, "grad_norm": 0.16077626109792695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61784167740, "tokens_per_second": 152017.29884691516, "data_wait_s": 1.5213998267427087e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 422200, "loss": 0.5677908658981323, "lr": 0.0003, "grad_norm": 0.14784672898380272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61798737904, "tokens_per_second": 150225.59618138464, "data_wait_s": 1.3166001735953614e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ae8564b4ed22cbf13aa7936c783e9741e21172cd6bb64283a53238173a76bc7
-size 533580
+oid sha256:f974b871027d0d1cc361950ae98513d8bcd9a6063ad0c88e1bf9865c03e58688
+size 535698