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": 75083.22647365948,
-      "required_pages_per_second_upper_estimate": 0.006508023774039084
+      "required_token_positions_per_second": 75184.84418691174,
+      "required_pages_per_second_upper_estimate": 0.006516831739876049
     },
     {
       "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": 9385.403309207435,
-      "required_pages_per_second_upper_estimate": 0.003743977704327204
+      "required_token_positions_per_second": 9398.105523363967,
+      "required_pages_per_second_upper_estimate": 0.0037490448074692707
     },
     {
       "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": 18770.80661841487,
-      "required_pages_per_second_upper_estimate": 0.004342295756256817
+      "required_token_positions_per_second": 18796.211046727934,
+      "required_pages_per_second_upper_estimate": 0.004348172623644377
     },
     {
       "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": 28156.2099276223,
-      "required_pages_per_second_upper_estimate": 0.001714565460375437
+      "required_token_positions_per_second": 28194.316570091905,
+      "required_pages_per_second_upper_estimate": 0.001716885954971734
     },
     {
       "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": 28156.2099276223,
-      "required_pages_per_second_upper_estimate": 0.0021675429816059143
+      "required_token_positions_per_second": 28194.316570091905,
+      "required_pages_per_second_upper_estimate": 0.0021704765364291626
     },
     {
       "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": 28156.2099276223,
-      "required_pages_per_second_upper_estimate": 0.34960589452825785
+      "required_token_positions_per_second": 28194.316570091905,
+      "required_pages_per_second_upper_estimate": 0.35007905149300206
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.21582327844530624,
-    "producer_p95_seconds": 0.16875636799977656,
-    "gpu_update_seconds": 0.9819503431417547,
-    "producer_batches_per_second": 4.633420487370731,
-    "gpu_batches_per_second": 1.0183814354608762,
-    "producer_headroom": 4.549788837493726,
+    "producer_batch_seconds": 0.06624915010156229,
+    "producer_p95_seconds": 0.17551731200001086,
+    "gpu_update_seconds": 0.9806231667742772,
+    "producer_batches_per_second": 15.09453326521117,
+    "gpu_batches_per_second": 1.0197597139066805,
+    "producer_headroom": 14.802049011511048,
     "sustainable": true,
-    "buffer_seconds": 60.88092127478879,
+    "buffer_seconds": 60.79863634000519,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 505538.93940051243,
-    "upload_bytes_per_second": 5785431.032807307,
+    "generation_bytes_per_second": 519587.71988912043,
+    "upload_bytes_per_second": 5929901.515746337,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.47688805114298,
+    "estimated_full_shards_per_hour": 111.49142930512629,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 6091488,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2760612950005452,
+      "bytes": 13277578,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -43,3 +43,4 @@
 {"step": 86200, "loss": 0.6315164659172297, "lr": 0.0003, "grad_norm": 0.1346323412636943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12614745751, "tokens_per_second": 147905.91825621613, "data_wait_s": 1.155600057245465e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 86300, "loss": 0.6972997039556503, "lr": 0.0003, "grad_norm": 0.15334566851523973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12629443000, "tokens_per_second": 149320.8702750207, "data_wait_s": 1.1412000276322942e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 86400, "loss": 0.6343935448676348, "lr": 0.0003, "grad_norm": 0.14075740858816985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12644099192, "tokens_per_second": 147445.3847662837, "data_wait_s": 1.1063000783906318e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 86500, "loss": 0.631697554141283, "lr": 0.0003, "grad_norm": 0.13317855507861112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12658762758, "tokens_per_second": 148322.60072593484, "data_wait_s": 1.1066000297432765e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eda9b73e55fcba4bccb70c629f3b57556b19695b5cc49aa29eb0006af930fcdd
-size 96258
+oid sha256:3d410b71d34e97a82b20f2df0d6a57d8b8ad2fd969eeef0dfbb52d4bbd320eaf
+size 98376