Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75995.98583923354,
-      "required_pages_per_second_upper_estimate": 0.006587139442479581
+      "required_token_positions_per_second": 75629.65283278764,
+      "required_pages_per_second_upper_estimate": 0.0065553866259433664
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9499.498229904193,
-      "required_pages_per_second_upper_estimate": 0.0037894918740642223
+      "required_token_positions_per_second": 9453.706604098456,
+      "required_pages_per_second_upper_estimate": 0.0037712249098844964
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18998.996459808386,
-      "required_pages_per_second_upper_estimate": 0.004395083460059163
+      "required_token_positions_per_second": 18907.41320819691,
+      "required_pages_per_second_upper_estimate": 0.004373897286609025
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28498.49468971258,
-      "required_pages_per_second_upper_estimate": 0.0017354088065573772
+      "required_token_positions_per_second": 28361.119812295365,
+      "required_pages_per_second_upper_estimate": 0.0017270433972729482
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28498.49468971258,
-      "required_pages_per_second_upper_estimate": 0.002193893010096489
+      "required_token_positions_per_second": 28361.119812295365,
+      "required_pages_per_second_upper_estimate": 0.002183317511754912
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28498.49468971258,
-      "required_pages_per_second_upper_estimate": 0.3538559257200117
+      "required_token_positions_per_second": 28361.119812295365,
+      "required_pages_per_second_upper_estimate": 0.3521501895066288
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.11424906661733303,
-    "producer_p95_seconds": 0.3757681510032853,
-    "gpu_update_seconds": 0.970156504791827,
-    "producer_batches_per_second": 8.752806737138695,
-    "gpu_batches_per_second": 1.0307615266823127,
-    "producer_headroom": 8.49159239122083,
+    "producer_batch_seconds": 0.099067899195461,
+    "producer_p95_seconds": 0.37787702799687395,
+    "gpu_update_seconds": 0.9748557244207894,
+    "producer_batches_per_second": 10.094087066760139,
+    "gpu_batches_per_second": 1.0257928172849886,
+    "producer_headroom": 9.840278559832976,
     "sustainable": true,
-    "buffer_seconds": 60.149703297093275,
+    "buffer_seconds": 60.441054914088944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381039.3716500027,
-    "upload_bytes_per_second": 5380083.113741708,
+    "generation_bytes_per_second": 380869.328477948,
+    "upload_bytes_per_second": 5263316.415894249,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.76217901349126,
+    "estimated_full_shards_per_hour": 81.72569170717077,
     "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": 1.3453164509992348,
-      "bytes": 11716150,
-      "pages": 31,
-      "shards": 4
-    },
     {
       "seconds": 0.8494216189974395,
       "bytes": 1181935,
@@ -562,6 +556,12 @@
       "bytes": 7506454,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0263809830066748,
+      "bytes": 924946,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -380,3 +380,4 @@
 {"step": 393000, "loss": 0.5575108397752047, "lr": 0.0003, "grad_norm": 0.14547461456295072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57524412536, "tokens_per_second": 148835.09593212197, "data_wait_s": 8.709001122042537e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 393100, "loss": 0.6420015506446362, "lr": 0.0003, "grad_norm": 0.1850644812418594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57539130193, "tokens_per_second": 149961.80861544103, "data_wait_s": 9.725001291371882e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 393200, "loss": 0.6698104962706566, "lr": 0.0003, "grad_norm": 0.18173146493014478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57553791015, "tokens_per_second": 149230.4514493089, "data_wait_s": 9.944997145794332e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 393300, "loss": 0.5766168162226677, "lr": 0.0003, "grad_norm": 0.16535435980387686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57568269703, "tokens_per_second": 146431.22659633058, "data_wait_s": 1.585899735800922e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8b7f3cf858b163b7f782d215b654d01e51422a38d2743262a8409bb83536219b
-size 814722
+oid sha256:7ac190699c8bfcb1bdffe7c413f8e1fc5e83f1e5cd71f55c0a1c0c88972ac87f
+size 816840