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": 74628.74988014875,
-      "required_pages_per_second_upper_estimate": 0.006468630894773969
+      "required_token_positions_per_second": 74950.1127229684,
+      "required_pages_per_second_upper_estimate": 0.006496485811502898
     },
     {
       "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": 9328.593735018594,
-      "required_pages_per_second_upper_estimate": 0.0037213155158044497
+      "required_token_positions_per_second": 9368.76409037105,
+      "required_pages_per_second_upper_estimate": 0.003737340071154879
     },
     {
       "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": 18657.187470037188,
-      "required_pages_per_second_upper_estimate": 0.004316011965908357
+      "required_token_positions_per_second": 18737.5281807421,
+      "required_pages_per_second_upper_estimate": 0.004334597375381722
     },
     {
       "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": 27985.78120505578,
-      "required_pages_per_second_upper_estimate": 0.001704187245341538
+      "required_token_positions_per_second": 28106.29227111315,
+      "required_pages_per_second_upper_estimate": 0.0017115257369917328
     },
     {
       "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": 27985.78120505578,
-      "required_pages_per_second_upper_estimate": 0.0021544229067658447
+      "required_token_positions_per_second": 28106.29227111315,
+      "required_pages_per_second_upper_estimate": 0.002163700182226923
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27985.78120505578,
-      "required_pages_per_second_upper_estimate": 0.34748974018222406
+      "required_token_positions_per_second": 28106.29227111315,
+      "required_pages_per_second_upper_estimate": 0.3489860842980636
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.34919488444531765,
-    "producer_p95_seconds": 0.17560051700002077,
-    "gpu_update_seconds": 0.9879302563476499,
-    "producer_batches_per_second": 2.8637303825010516,
-    "gpu_batches_per_second": 1.012217202150455,
-    "producer_headroom": 2.829165890894817,
+    "producer_batch_seconds": 0.06901966136717519,
+    "producer_p95_seconds": 0.17981109399988782,
+    "gpu_update_seconds": 0.9836943177459173,
+    "producer_batches_per_second": 14.488625127847794,
+    "gpu_batches_per_second": 1.0165759646669976,
+    "producer_headroom": 14.252378210214589,
     "sustainable": true,
-    "buffer_seconds": 60.26374563720664,
+    "buffer_seconds": 60.00535338250096,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290987.732077614,
-    "upload_bytes_per_second": 3425789.0439186757,
+    "generation_bytes_per_second": 362453.50531179394,
+    "upload_bytes_per_second": 5437030.695865118,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.4391934561378,
+    "estimated_full_shards_per_hour": 77.7740847541367,
     "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": [
@@ -190,6 +190,12 @@
       "bytes": 9815303,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.441364390000217,
+      "bytes": 13613764,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -31,3 +31,4 @@
 {"step": 198100, "loss": 0.5864313617348671, "lr": 0.0003, "grad_norm": 0.13070283264529284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28994197011, "tokens_per_second": 148100.4656571597, "data_wait_s": 1.8413000361761078e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 198200, "loss": 0.6315398290753365, "lr": 0.0003, "grad_norm": 0.15816601913565825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29008862048, "tokens_per_second": 147508.96141309646, "data_wait_s": 2.0446999769774266e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 198300, "loss": 0.5880254246294498, "lr": 0.0003, "grad_norm": 0.13477129061986676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29023536881, "tokens_per_second": 148109.66909596414, "data_wait_s": 1.2978000086150132e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 198400, "loss": 0.6215808764100075, "lr": 0.0003, "grad_norm": 0.1683647299050308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29038047497, "tokens_per_second": 146338.90368490835, "data_wait_s": 1.6541999684704933e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3f62ae795b51f4c3695ed5e2a22877c27290d36201b39f1d6b064d122e855fd
-size 70059
+oid sha256:761a30d59c129ceae09b6334d237cc1bff37160abf8923b68cbcd6a25ec79afd
+size 72177