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": 75955.3176128496,
-      "required_pages_per_second_upper_estimate": 0.005254765113075015
+      "required_token_positions_per_second": 77190.71732742061,
+      "required_pages_per_second_upper_estimate": 0.005340232931851306
     },
     {
       "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": 9494.4147016062,
-      "required_pages_per_second_upper_estimate": 0.003975113169637976
+      "required_token_positions_per_second": 9648.839665927577,
+      "required_pages_per_second_upper_estimate": 0.004039767677439382
     },
     {
       "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": 18988.8294032124,
-      "required_pages_per_second_upper_estimate": 0.003954872395582516
+      "required_token_positions_per_second": 19297.679331855154,
+      "required_pages_per_second_upper_estimate": 0.004019197690798461
     },
     {
       "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": 28483.244104818597,
-      "required_pages_per_second_upper_estimate": 0.0017075269926883603
+      "required_token_positions_per_second": 28946.518997782732,
+      "required_pages_per_second_upper_estimate": 0.0017352996151417572
     },
     {
       "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": 28483.244104818597,
-      "required_pages_per_second_upper_estimate": 0.002214446944599308
+      "required_token_positions_per_second": 28946.518997782732,
+      "required_pages_per_second_upper_estimate": 0.0022504645297963697
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28483.244104818597,
-      "required_pages_per_second_upper_estimate": 0.365352472451849
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28946.518997782732,
+      "required_pages_per_second_upper_estimate": 0.3712948653529679
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3101006119218823,
-    "producer_p95_seconds": 0.3071302360003756,
-    "gpu_update_seconds": 0.9706759489282579,
-    "producer_batches_per_second": 3.2247598410154406,
-    "gpu_batches_per_second": 1.0302099285597004,
-    "producer_headroom": 3.130196818743401,
+    "producer_batch_seconds": 0.17675065844528604,
+    "producer_p95_seconds": 0.29572768499929225,
+    "gpu_update_seconds": 0.9551407546488683,
+    "producer_batches_per_second": 5.657687551469883,
+    "gpu_batches_per_second": 1.0469661095841556,
+    "producer_headroom": 5.403887957478452,
     "sustainable": true,
-    "buffer_seconds": 60.181908833551994,
+    "buffer_seconds": 60.1738675428787,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 525135.3264741183,
-    "upload_bytes_per_second": 6072452.474192158,
+    "generation_bytes_per_second": 524405.7101825735,
+    "upload_bytes_per_second": 6133694.282782401,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.68181653659498,
+    "estimated_full_shards_per_hour": 112.52525786502747,
     "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": 0.9707876079992275,
-      "bytes": 899756,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.1750758769994718,
       "bytes": 10630556,
@@ -562,6 +556,12 @@
       "bytes": 12027088,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.2329098789996351,
+      "bytes": 6852339,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -298,3 +298,4 @@
 {"step": 514800, "loss": 0.5757641196250916, "lr": 0.0003, "grad_norm": 0.18948936053082832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75349244454, "tokens_per_second": 152224.3600448378, "data_wait_s": 1.2065000191796571e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 514900, "loss": 0.5893741846084595, "lr": 0.0003, "grad_norm": 0.1534901180199586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75363913078, "tokens_per_second": 152991.34452814868, "data_wait_s": 2.158999996026978e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 515000, "loss": 0.6022728681564331, "lr": 0.0003, "grad_norm": 0.16161594023367803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75378632707, "tokens_per_second": 153146.02722571362, "data_wait_s": 1.647099998081103e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 515100, "loss": 0.6161287426948547, "lr": 0.0003, "grad_norm": 0.18326037244278115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75393312170, "tokens_per_second": 145675.3699440447, "data_wait_s": 1.7000998923322186e-05, "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 sha256:16085e31b7a9d50d28f4ef25a1a8c0aa0e9fb0a31ee3e61becdb2fc62c1a3185
-size 640263
+oid sha256:97141224750d9d4f925a97c57bb47aa488217692650156cb0f428399b1d2cf3b
+size 642381