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": 75652.3251866741,
-      "required_pages_per_second_upper_estimate": 0.006557351808115689
+      "required_token_positions_per_second": 75836.29419891983,
+      "required_pages_per_second_upper_estimate": 0.00657329777583195
     },
     {
       "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": 9456.540648334263,
-      "required_pages_per_second_upper_estimate": 0.0037723554525028974
+      "required_token_positions_per_second": 9479.536774864979,
+      "required_pages_per_second_upper_estimate": 0.003781528951198731
     },
     {
       "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": 18913.081296668526,
-      "required_pages_per_second_upper_estimate": 0.0043752084991219835
+      "required_token_positions_per_second": 18959.073549729957,
+      "required_pages_per_second_upper_estimate": 0.004385847997431738
     },
     {
       "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": 28369.62194500279,
-      "required_pages_per_second_upper_estimate": 0.0017275611325475873
+      "required_token_positions_per_second": 28438.610324594934,
+      "required_pages_per_second_upper_estimate": 0.0017317621629106947
     },
     {
       "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": 28369.62194500279,
-      "required_pages_per_second_upper_estimate": 0.002183972029466139
+      "required_token_positions_per_second": 28438.610324594934,
+      "required_pages_per_second_upper_estimate": 0.0021892829458992
     },
     {
       "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": 28369.62194500279,
-      "required_pages_per_second_upper_estimate": 0.3522557575400473
+      "required_token_positions_per_second": 28438.610324594934,
+      "required_pages_per_second_upper_estimate": 0.3531123623253279
     }
   ],
   "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": 0.42316093439845304,
-    "producer_p95_seconds": 0.20831448300123157,
-    "gpu_update_seconds": 0.9745635685099462,
-    "producer_batches_per_second": 2.363167104311167,
-    "gpu_batches_per_second": 1.0261003307654366,
-    "producer_headroom": 2.303056566162807,
+    "producer_batch_seconds": 0.1140174884921521,
+    "producer_p95_seconds": 0.25114188400038984,
+    "gpu_update_seconds": 0.9721994037130858,
+    "producer_batches_per_second": 8.7705843482847,
+    "gpu_batches_per_second": 1.0285955701893423,
+    "producer_headroom": 8.526756873617707,
     "sustainable": true,
-    "buffer_seconds": 60.42294124761666,
+    "buffer_seconds": 60.27636303021132,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 205263.4972713083,
-    "upload_bytes_per_second": 6555182.658091796,
+    "generation_bytes_per_second": 221987.01268646927,
+    "upload_bytes_per_second": 6385706.1185081825,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 44.0447682247585,
+    "estimated_full_shards_per_hour": 47.633245329337676,
     "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": 1.3273477019993152,
-      "bytes": 12043048,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 0.7578286200005095,
       "bytes": 992201,
@@ -562,6 +556,12 @@
       "bytes": 995690,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.5069925719999446,
+      "bytes": 1775692,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -133,3 +133,4 @@
 {"step": 13300, "loss": 0.6555228605866432, "lr": 0.0003, "grad_norm": 0.15461106607022207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1945814844, "tokens_per_second": 148214.7037890147, "data_wait_s": 1.1725998774636537e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 13400, "loss": 0.7065081484615803, "lr": 0.0003, "grad_norm": 0.15345629303977382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1960433967, "tokens_per_second": 149251.74296571227, "data_wait_s": 1.3222001143731177e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 13500, "loss": 0.6960186250507832, "lr": 0.0003, "grad_norm": 0.145830692970031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1975092483, "tokens_per_second": 147651.45616285558, "data_wait_s": 9.116000001085922e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 13600, "loss": 0.6929355002939701, "lr": 0.0003, "grad_norm": 0.14627108368837513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1989608705, "tokens_per_second": 138886.82479895413, "data_wait_s": 1.1251000614720397e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:61d7bae9ab1d611c84b4bf7ca5579578084e4f0fe5c92803199c4fe6d41ae3ee
-size 282877
+oid sha256:c4f96e909b1b0533741d64a7366c1d70571564b5799bb86b9bd6d56dbb050ded
+size 284960