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": 76547.62205692276,
-      "required_pages_per_second_upper_estimate": 0.005295742108851653
+      "required_token_positions_per_second": 76555.4465215218,
+      "required_pages_per_second_upper_estimate": 0.005296283423467874
     },
     {
       "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": 9568.452757115345,
-      "required_pages_per_second_upper_estimate": 0.004006111357389245
+      "required_token_positions_per_second": 9569.430815190224,
+      "required_pages_per_second_upper_estimate": 0.004006520849881022
     },
     {
       "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": 19136.90551423069,
-      "required_pages_per_second_upper_estimate": 0.003985712744477976
+      "required_token_positions_per_second": 19138.86163038045,
+      "required_pages_per_second_upper_estimate": 0.003986120151885719
     },
     {
       "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": 28705.358271346035,
-      "required_pages_per_second_upper_estimate": 0.0017208423978229853
+      "required_token_positions_per_second": 28708.292445570674,
+      "required_pages_per_second_upper_estimate": 0.0017210182970875277
     },
     {
       "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": 28705.358271346035,
-      "required_pages_per_second_upper_estimate": 0.002231715344069848
+      "required_token_positions_per_second": 28708.292445570674,
+      "required_pages_per_second_upper_estimate": 0.0022319434632097454
     },
     {
       "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": 28705.358271346035,
-      "required_pages_per_second_upper_estimate": 0.3682015144924518
+      "required_token_positions_per_second": 28708.292445570674,
+      "required_pages_per_second_upper_estimate": 0.36823915092893467
     }
   ],
   "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": 0.4238600857108281,
+    "producer_batch_seconds": 0.07847373991387485,
     "producer_p95_seconds": 0.2640660069992009,
-    "gpu_update_seconds": 0.963165125432296,
-    "producer_batches_per_second": 2.359269093061606,
-    "gpu_batches_per_second": 1.038243571735606,
-    "producer_headroom": 2.272365711947221,
+    "gpu_update_seconds": 0.9630666836914481,
+    "producer_batches_per_second": 12.743116373674846,
+    "gpu_batches_per_second": 1.0383496978288003,
+    "producer_headroom": 12.272470825889227,
     "sustainable": true,
-    "buffer_seconds": 60.67940290223465,
+    "buffer_seconds": 60.67320107256123,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279524.8085141819,
-    "upload_bytes_per_second": 5429886.592322225,
+    "generation_bytes_per_second": 303848.44854318624,
+    "upload_bytes_per_second": 5455756.59824723,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.97951690262882,
+    "estimated_full_shards_per_hour": 65.19880382749263,
     "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": 1.3860260590008693,
-      "bytes": 10598757,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.0763569690025179,
       "bytes": 2341572,
@@ -562,6 +556,12 @@
       "bytes": 1024576,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0973008010005287,
+      "bytes": 11080849,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -270,3 +270,4 @@
 {"step": 747000, "loss": 0.49867814779281616, "lr": 5e-05, "grad_norm": 0.22463947518871102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109340651401, "tokens_per_second": 150860.76193270652, "data_wait_s": 8.994000381790102e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 747100, "loss": 0.47659775614738464, "lr": 5e-05, "grad_norm": 0.2503915523096495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109355448706, "tokens_per_second": 153298.50200161533, "data_wait_s": 1.634599902899936e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 747200, "loss": 0.5707584619522095, "lr": 5e-05, "grad_norm": 0.2587037895014516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109370052261, "tokens_per_second": 150716.58541321536, "data_wait_s": 1.489500209572725e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 747300, "loss": 0.5461853742599487, "lr": 5e-05, "grad_norm": 0.23208668365497262, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109384491561, "tokens_per_second": 148383.02233741563, "data_wait_s": 1.5613000869052485e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c25e25d82173a0804700737e5b4a62743a9ea4bb8456846264a747c378c7ee9
-size 580176
+oid sha256:7e8d03a010f2f153f33a66fb3f7301d6871c06b9e0ea75470de2fefc3be9b7dd
+size 582294