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": 76703.71848483548,
-      "required_pages_per_second_upper_estimate": 0.005306541221928269
+      "required_token_positions_per_second": 76370.76949295575,
+      "required_pages_per_second_upper_estimate": 0.005283507038121932
     },
     {
       "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": 9587.964810604435,
-      "required_pages_per_second_upper_estimate": 0.004014280646727117
+      "required_token_positions_per_second": 9546.346186619468,
+      "required_pages_per_second_upper_estimate": 0.003996855797960308
     },
     {
       "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": 19175.92962120887,
-      "required_pages_per_second_upper_estimate": 0.003993840436826625
+      "required_token_positions_per_second": 19092.692373238937,
+      "required_pages_per_second_upper_estimate": 0.003976504313188331
     },
     {
       "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": 28763.894431813304,
-      "required_pages_per_second_upper_estimate": 0.0017243515512634562
+      "required_token_positions_per_second": 28639.038559858407,
+      "required_pages_per_second_upper_estimate": 0.0017168666323836381
     },
     {
       "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": 28763.894431813304,
-      "required_pages_per_second_upper_estimate": 0.002236266273072819
+      "required_token_positions_per_second": 28639.038559858407,
+      "required_pages_per_second_upper_estimate": 0.002226559278211267
     },
     {
       "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": 28763.894431813304,
-      "required_pages_per_second_upper_estimate": 0.3689523535076936
+      "required_token_positions_per_second": 28639.038559858407,
+      "required_pages_per_second_upper_estimate": 0.36735083644204675
     }
   ],
   "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": 0.12856291852338586,
-    "producer_p95_seconds": 0.32105255799979204,
-    "gpu_update_seconds": 0.9612050296437221,
-    "producer_batches_per_second": 7.778292617230044,
-    "gpu_batches_per_second": 1.0403607650395437,
-    "producer_headroom": 7.47653398572215,
+    "producer_batch_seconds": 0.08745149480452596,
+    "producer_p95_seconds": 0.3146036760008428,
+    "gpu_update_seconds": 0.9653955366627606,
+    "producer_batches_per_second": 11.434910314972067,
+    "gpu_batches_per_second": 1.0358448553189528,
+    "producer_headroom": 11.039211380212993,
     "sustainable": true,
-    "buffer_seconds": 60.555916867554494,
+    "buffer_seconds": 60.819918809753915,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 520231.5002343338,
-    "upload_bytes_per_second": 5115899.13953586,
+    "generation_bytes_per_second": 553817.9680738361,
+    "upload_bytes_per_second": 5070954.374711999,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.62956958076964,
+    "estimated_full_shards_per_hour": 118.83644372617064,
     "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": 1.4267704899975797,
-      "bytes": 7513990,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0001321140007349,
       "bytes": 5054764,
@@ -562,6 +556,12 @@
       "bytes": 7577948,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.8521174329980568,
+      "bytes": 865962,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -271,3 +271,4 @@
 {"step": 631500, "loss": 0.5649096369743347, "lr": 0.0001, "grad_norm": 0.17203733025526274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92435093571, "tokens_per_second": 152675.37007872996, "data_wait_s": 1.105699993786402e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 631600, "loss": 0.49820688366889954, "lr": 0.0001, "grad_norm": 0.16837591631471038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92449576074, "tokens_per_second": 149642.83750909698, "data_wait_s": 1.789499947335571e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 631700, "loss": 0.5200865268707275, "lr": 0.0001, "grad_norm": 0.16134545844521925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92464076764, "tokens_per_second": 150018.86339453675, "data_wait_s": 1.6152000171132386e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 631800, "loss": 0.49992790818214417, "lr": 0.0001, "grad_norm": 0.1781931082507836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92478667705, "tokens_per_second": 150913.55551660372, "data_wait_s": 1.8184000509791076e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9adb2d37b345ad1d45b3f19db324acda2fe995227023d5a476526ea78e84fe78
-size 583077
+oid sha256:88740d3c798deb5414a90f53bf92817d36f62121d77d5057496490d042421234
+size 585195