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": 76353.1900979817,
-      "required_pages_per_second_upper_estimate": 0.005282290854787811
+      "required_token_positions_per_second": 77200.01258345156,
+      "required_pages_per_second_upper_estimate": 0.005340875999231493
     },
     {
       "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": 9544.148762247713,
-      "required_pages_per_second_upper_estimate": 0.003995935782263292
+      "required_token_positions_per_second": 9650.001572931445,
+      "required_pages_per_second_upper_estimate": 0.004040254143638525
     },
     {
       "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": 19088.297524495425,
-      "required_pages_per_second_upper_estimate": 0.003975588982095013
+      "required_token_positions_per_second": 19300.00314586289,
+      "required_pages_per_second_upper_estimate": 0.0040196816799731765
     },
     {
       "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": 28632.446286743136,
-      "required_pages_per_second_upper_estimate": 0.0017164714356761957
+      "required_token_positions_per_second": 28950.004718794335,
+      "required_pages_per_second_upper_estimate": 0.001735508578794015
     },
     {
       "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": 28632.446286743136,
-      "required_pages_per_second_upper_estimate": 0.0022260467579729
+      "required_token_positions_per_second": 28950.004718794335,
+      "required_pages_per_second_upper_estimate": 0.0022507355292729556
     },
     {
       "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": 28632.446286743136,
-      "required_pages_per_second_upper_estimate": 0.36726627784075544
+      "required_token_positions_per_second": 28950.004718794335,
+      "required_pages_per_second_upper_estimate": 0.3713395764394291
     }
   ],
   "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": 0.06556022986720933,
-    "producer_p95_seconds": 0.17263578999973106,
-    "gpu_update_seconds": 0.9656178072636799,
-    "producer_batches_per_second": 15.253149691901264,
-    "gpu_batches_per_second": 1.0356064195147257,
-    "producer_headroom": 14.728712959358372,
+    "producer_batch_seconds": 0.04976373940626644,
+    "producer_p95_seconds": 0.1702074619997802,
+    "gpu_update_seconds": 0.9550257510683902,
+    "producer_batches_per_second": 20.094952910111818,
+    "gpu_batches_per_second": 1.0470921845628738,
+    "producer_headroom": 19.191197495663474,
     "sustainable": true,
-    "buffer_seconds": 60.83392185761183,
+    "buffer_seconds": 60.16662231730858,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288426.44305521855,
-    "upload_bytes_per_second": 5481240.966060104,
+    "generation_bytes_per_second": 317438.5499692344,
+    "upload_bytes_per_second": 5504378.909206007,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.889600455688644,
+    "estimated_full_shards_per_hour": 68.11492323215269,
     "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": 1.022603193000009,
-      "bytes": 47630,
-      "pages": 1,
-      "shards": 1
-    },
     {
       "seconds": 0.8258607079999933,
       "bytes": 145464,
@@ -562,6 +556,12 @@
       "bytes": 7619131,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.8663683109998601,
+      "bytes": 939323,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -76,3 +76,4 @@
 {"step": 564700, "loss": 0.5635547041893005, "lr": 0.0003, "grad_norm": 0.17389888051443375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82654803122, "tokens_per_second": 152585.3997248335, "data_wait_s": 1.554000027681468e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 564800, "loss": 0.5848207473754883, "lr": 0.0003, "grad_norm": 0.15228925910955995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82669492000, "tokens_per_second": 153143.66669734853, "data_wait_s": 1.4373999874806032e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 564900, "loss": 0.568290114402771, "lr": 0.0003, "grad_norm": 0.14954964407563132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82684113311, "tokens_per_second": 152003.60723416286, "data_wait_s": 1.2790999790013302e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 565000, "loss": 0.5508708357810974, "lr": 0.0003, "grad_norm": 0.1489226743175592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82698757298, "tokens_per_second": 152888.4508686426, "data_wait_s": 1.3930000022810418e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2a3c16abe79d4eb76d2e29765ec5beb1888b1805669d7a6e4212003d174fd5d4
-size 166152
+oid sha256:ce42931f8d0fc365d65d7f37962600d0e95b4c9076453387c01111eb6a49b5ef
+size 168270