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": 76731.92156450354,
-      "required_pages_per_second_upper_estimate": 0.005308492376419866
+      "required_token_positions_per_second": 77276.97742404194,
+      "required_pages_per_second_upper_estimate": 0.005346200605486569
     },
     {
       "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": 9591.490195562943,
-      "required_pages_per_second_upper_estimate": 0.0040157566517908345
+      "required_token_positions_per_second": 9659.622178005242,
+      "required_pages_per_second_upper_estimate": 0.0040442820900818445
     },
     {
       "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": 19182.980391125886,
-      "required_pages_per_second_upper_estimate": 0.003995308926259032
+      "required_token_positions_per_second": 19319.244356010484,
+      "required_pages_per_second_upper_estimate": 0.004023689116622086
     },
     {
       "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": 28774.470586688833,
-      "required_pages_per_second_upper_estimate": 0.0017249855755993907
+      "required_token_positions_per_second": 28978.866534015724,
+      "required_pages_per_second_upper_estimate": 0.0017372388005470973
     },
     {
       "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": 28774.470586688833,
-      "required_pages_per_second_upper_estimate": 0.0022370885225948024
+      "required_token_positions_per_second": 28978.866534015724,
+      "required_pages_per_second_upper_estimate": 0.00225297940845671
     },
     {
       "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": 28774.470586688833,
-      "required_pages_per_second_upper_estimate": 0.36908801306664657
+      "required_token_positions_per_second": 28978.866534015724,
+      "required_pages_per_second_upper_estimate": 0.37170978481568634
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.23257930135969218,
-    "producer_p95_seconds": 0.3252726580030867,
-    "gpu_update_seconds": 0.9608517354543461,
-    "producer_batches_per_second": 4.299608753461102,
-    "gpu_batches_per_second": 1.040743293789382,
-    "producer_headroom": 4.131286532537797,
+    "producer_batch_seconds": 0.1328595010702429,
+    "producer_p95_seconds": 0.3357842309997068,
+    "gpu_update_seconds": 0.9540745828532134,
+    "producer_batches_per_second": 7.5267481207181355,
+    "gpu_batches_per_second": 1.0481360870231382,
+    "producer_headroom": 7.181079073515364,
     "sustainable": true,
-    "buffer_seconds": 60.5336593336238,
+    "buffer_seconds": 60.106698719752444,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 253439.59573993727,
-    "upload_bytes_per_second": 4979433.471407232,
+    "generation_bytes_per_second": 254435.64449582653,
+    "upload_bytes_per_second": 4973390.621708597,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.38223747395123,
+    "estimated_full_shards_per_hour": 54.59596635013673,
     "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": 1.506981171998632,
-      "bytes": 11206977,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 0.8770163509980193,
       "bytes": 891965,
@@ -562,6 +556,12 @@
       "bytes": 6049937,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.4160137470025802,
+      "bytes": 10237474,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -410,3 +410,4 @@
 {"step": 926000, "loss": 0.5182063579559326, "lr": 3e-05, "grad_norm": 0.3363843247925149, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135540055118, "tokens_per_second": 119356.26496268879, "data_wait_s": 1.3778000720776618e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 926100, "loss": 0.5734084248542786, "lr": 3e-05, "grad_norm": 0.3732993021334413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135554837206, "tokens_per_second": 153593.45358336106, "data_wait_s": 1.2283999240025878e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 926200, "loss": 0.584020733833313, "lr": 3e-05, "grad_norm": 0.3458658570927081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135569601061, "tokens_per_second": 154412.24199963108, "data_wait_s": 1.1240997991990298e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 926300, "loss": 0.5263802409172058, "lr": 3e-05, "grad_norm": 0.3582380577863063, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135584278242, "tokens_per_second": 153598.82214552615, "data_wait_s": 1.4899997040629387e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1acfb7f5268930f178d4ca475fed2f0ee0227213fb735f383b078f9ef52b1fb5
-size 879045
+oid sha256:3a9424ef52cbef7418e216852a9dfc23345e5db1e35a5fa0a50295f2d1125d95
+size 881163