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": 76101.1899025512,
-      "required_pages_per_second_upper_estimate": 0.006596258264050694
+      "required_token_positions_per_second": 75885.54792210295,
+      "required_pages_per_second_upper_estimate": 0.006577566963725043
     },
     {
       "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": 9512.6487378189,
-      "required_pages_per_second_upper_estimate": 0.0037947378082890137
+      "required_token_positions_per_second": 9485.69349026287,
+      "required_pages_per_second_upper_estimate": 0.0037839849570220478
     },
     {
       "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": 19025.2974756378,
-      "required_pages_per_second_upper_estimate": 0.004401167737189228
+      "required_token_positions_per_second": 18971.38698052574,
+      "required_pages_per_second_upper_estimate": 0.004388696492937361
     },
     {
       "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": 28537.946213456697,
-      "required_pages_per_second_upper_estimate": 0.0017378111973672455
+      "required_token_positions_per_second": 28457.08047078861,
+      "required_pages_per_second_upper_estimate": 0.0017328868979085215
     },
     {
       "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": 28537.946213456697,
-      "required_pages_per_second_upper_estimate": 0.002196930097603118
+      "required_token_positions_per_second": 28457.08047078861,
+      "required_pages_per_second_upper_estimate": 0.002190704828881826
     },
     {
       "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": 28537.946213456697,
-      "required_pages_per_second_upper_estimate": 0.35434578160915725
+      "required_token_positions_per_second": 28457.08047078861,
+      "required_pages_per_second_upper_estimate": 0.35334169972545054
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.07675964681243386,
-    "producer_p95_seconds": 0.17961427900081617,
-    "gpu_update_seconds": 0.9688153377681729,
-    "producer_batches_per_second": 13.027678494190463,
-    "gpu_batches_per_second": 1.0321884481140298,
-    "producer_headroom": 12.621414740684294,
+    "producer_batch_seconds": 0.21056683703898216,
+    "producer_p95_seconds": 0.1795413069994538,
+    "gpu_update_seconds": 0.9715683950214381,
+    "producer_batches_per_second": 4.749085915247282,
+    "gpu_batches_per_second": 1.0292636165649816,
+    "producer_headroom": 4.61406178049572,
     "sustainable": true,
-    "buffer_seconds": 60.066550941626716,
+    "buffer_seconds": 60.23724049132916,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 101,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 55,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 793581.97609403,
-    "upload_bytes_per_second": 3309682.6814441257,
+    "generation_bytes_per_second": 495513.670019019,
+    "upload_bytes_per_second": 4175948.7004452706,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.14210921342695,
+    "estimated_full_shards_per_hour": 106.32569861820151,
     "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": 7791997,
       "pages": 45,
       "shards": 3
+    },
+    {
+      "seconds": 0.999563274999673,
+      "bytes": 6224997,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -2,3 +2,4 @@
 {"step": 280400, "loss": 0.6441070791333914, "lr": 0.0003, "grad_norm": 0.15491111289274895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41042964989, "tokens_per_second": 151113.72276237543, "data_wait_s": 1.8066000848193653e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 280500, "loss": 0.6065780781209469, "lr": 0.0003, "grad_norm": 0.16653364277928817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41057650350, "tokens_per_second": 150594.33820475705, "data_wait_s": 1.4020999515196308e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 280600, "loss": 0.6250142902135849, "lr": 0.0003, "grad_norm": 0.17961906661422258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41072416030, "tokens_per_second": 151388.31445392434, "data_wait_s": 1.0661000487743877e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 280700, "loss": 0.6278036907315254, "lr": 0.0003, "grad_norm": 0.15855921155288702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41087038742, "tokens_per_second": 150171.5709545002, "data_wait_s": 1.1882999388035387e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aeb0f24910397fdf11cda0d92bfe45b8591774562d27ddf523512c5fd8d15924
-size 8637
+oid sha256:1ec02be7479c7d51b9c6caf1f3a7b63f43050762b04f0afea3cef00b70fd43e7
+size 10755