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": 76185.18029287754,
-      "required_pages_per_second_upper_estimate": 0.006603538338212484
+      "required_token_positions_per_second": 75632.79396915692,
+      "required_pages_per_second_upper_estimate": 0.006555658891682718
     },
     {
       "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": 9523.147536609693,
-      "required_pages_per_second_upper_estimate": 0.0037989259360976914
+      "required_token_positions_per_second": 9454.099246144615,
+      "required_pages_per_second_upper_estimate": 0.003771381540667231
     },
     {
       "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": 19046.295073219386,
-      "required_pages_per_second_upper_estimate": 0.004406025161844632
+      "required_token_positions_per_second": 18908.19849228923,
+      "required_pages_per_second_upper_estimate": 0.004374078948263256
     },
     {
       "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": 28569.44260982908,
-      "required_pages_per_second_upper_estimate": 0.0017397291626575029
+      "required_token_positions_per_second": 28362.29773843384,
+      "required_pages_per_second_upper_estimate": 0.0017271151267946804
     },
     {
       "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": 28569.44260982908,
-      "required_pages_per_second_upper_estimate": 0.0021993547773834694
+      "required_token_positions_per_second": 28362.29773843384,
+      "required_pages_per_second_upper_estimate": 0.00218340819176976
     },
     {
       "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": 28569.44260982908,
-      "required_pages_per_second_upper_estimate": 0.3547368614404445
+      "required_token_positions_per_second": 28362.29773843384,
+      "required_pages_per_second_upper_estimate": 0.3521648154070035
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.24357236725788312,
-    "producer_p95_seconds": 0.31968467899787356,
-    "gpu_update_seconds": 0.9677472668118466,
-    "producer_batches_per_second": 4.105556025332079,
-    "gpu_batches_per_second": 1.0333276406911558,
-    "producer_headroom": 3.973140622258028,
+    "producer_batch_seconds": 0.09901512860935213,
+    "producer_p95_seconds": 0.29387518700241344,
+    "gpu_update_seconds": 0.9748152372906693,
+    "producer_batches_per_second": 10.099466758714572,
+    "gpu_batches_per_second": 1.0258354216736778,
+    "producer_headroom": 9.845114084905571,
     "sustainable": true,
-    "buffer_seconds": 60.00033054233449,
+    "buffer_seconds": 60.438544712021496,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 540847.5994372981,
-    "upload_bytes_per_second": 5116793.03420257,
+    "generation_bytes_per_second": 482046.66169499245,
+    "upload_bytes_per_second": 4725125.335335737,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.05330455149848,
+    "estimated_full_shards_per_hour": 103.43599212777453,
     "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": 1.024946067998826,
-      "bytes": 5141639,
-      "pages": 39,
-      "shards": 2
-    },
     {
       "seconds": 1.2369695059969672,
       "bytes": 7463645,
@@ -562,6 +556,12 @@
       "bytes": 10255842,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 8.627843940001185,
+      "bytes": 8388512,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -244,3 +244,4 @@
 {"step": 64400, "loss": 0.6146517023444176, "lr": 0.0003, "grad_norm": 0.1410616037631765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9424838604, "tokens_per_second": 148309.56121770755, "data_wait_s": 9.709001460578293e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 64500, "loss": 0.6141596455127001, "lr": 0.0003, "grad_norm": 0.12958309178787517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9439627986, "tokens_per_second": 152102.54236346792, "data_wait_s": 9.973999112844467e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 64600, "loss": 0.6922388523817062, "lr": 0.0003, "grad_norm": 0.1990004472183067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9454299860, "tokens_per_second": 148879.53695223405, "data_wait_s": 8.094000804703683e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 64700, "loss": 0.6381540410220623, "lr": 0.0003, "grad_norm": 0.15739835779292288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9469111092, "tokens_per_second": 151236.28346579967, "data_wait_s": 8.705999789526686e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7142843a458ca4f480e38a43cf1ed7bad2b99c9fafb9d88965a04ad4dc0eb49
-size 524325
+oid sha256:dc7cc15efe77d5ec584709cc86818037aed628b73fd1902a3627f18f1545fbdd
+size 526443