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": 77200.01258345156,
-      "required_pages_per_second_upper_estimate": 0.005340875999231493
+      "required_token_positions_per_second": 77151.47398076436,
+      "required_pages_per_second_upper_estimate": 0.005337517986072517
     },
     {
       "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": 9650.001572931445,
-      "required_pages_per_second_upper_estimate": 0.004040254143638525
+      "required_token_positions_per_second": 9643.934247595545,
+      "required_pages_per_second_upper_estimate": 0.004037713881220544
     },
     {
       "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": 19300.00314586289,
-      "required_pages_per_second_upper_estimate": 0.0040196816799731765
+      "required_token_positions_per_second": 19287.86849519109,
+      "required_pages_per_second_upper_estimate": 0.004017154352250499
     },
     {
       "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": 28950.004718794335,
-      "required_pages_per_second_upper_estimate": 0.001735508578794015
+      "required_token_positions_per_second": 28931.802742786636,
+      "required_pages_per_second_upper_estimate": 0.0017344173981251616
     },
     {
       "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": 28950.004718794335,
-      "required_pages_per_second_upper_estimate": 0.0022507355292729556
+      "required_token_positions_per_second": 28931.802742786636,
+      "required_pages_per_second_upper_estimate": 0.002249320405700388
     },
     {
       "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": 28950.004718794335,
-      "required_pages_per_second_upper_estimate": 0.3713395764394291
+      "required_token_positions_per_second": 28931.802742786636,
+      "required_pages_per_second_upper_estimate": 0.37110610103496156
     }
   ],
   "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": 0.04976373940626644,
-    "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,
+    "producer_batch_seconds": 0.09874852467972062,
+    "producer_p95_seconds": 0.19699957800003176,
+    "gpu_update_seconds": 0.9556265900815075,
+    "producer_batches_per_second": 10.126733571396473,
+    "gpu_batches_per_second": 1.046433837629725,
+    "producer_headroom": 9.677375871497539,
     "sustainable": true,
-    "buffer_seconds": 60.16662231730858,
+    "buffer_seconds": 60.204475175134974,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317438.5499692344,
-    "upload_bytes_per_second": 5504378.909206007,
+    "generation_bytes_per_second": 317377.99469481764,
+    "upload_bytes_per_second": 5576105.771814248,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.11492323215269,
+    "estimated_full_shards_per_hour": 68.1019294799175,
     "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": 0.8258607079999933,
-      "bytes": 145464,
-      "pages": 4,
-      "shards": 1
-    },
     {
       "seconds": 1.1176107269999989,
       "bytes": 5541224,
@@ -562,6 +556,12 @@
       "bytes": 939323,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 0.8838954740003828,
+      "bytes": 5887982,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -77,3 +77,4 @@
 {"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}
+{"step": 565100, "loss": 0.6213895678520203, "lr": 0.0003, "grad_norm": 0.1489691330117435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82713498741, "tokens_per_second": 139122.28454347706, "data_wait_s": 1.5305000488297082e-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 sha256:7cb1667e35fc817dc071ebccb8e255101452f58214fddf4e1cf77306f98c72f7
-size 169053
+oid sha256:f1880235d314725eef74fcef14030f4e2cdf4fc23454805e371624a3e6a198fc
+size 171171