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": 75564.60575308664,
-      "required_pages_per_second_upper_estimate": 0.0065497485099352
+      "required_token_positions_per_second": 75707.47302353606,
+      "required_pages_per_second_upper_estimate": 0.0065621318828439695
     },
     {
       "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": 9445.57571913583,
-      "required_pages_per_second_upper_estimate": 0.0037679813783053416
+      "required_token_positions_per_second": 9463.434127942008,
+      "required_pages_per_second_upper_estimate": 0.0037751053645851316
     },
     {
       "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": 18891.15143827166,
-      "required_pages_per_second_upper_estimate": 0.004370135412334194
+      "required_token_positions_per_second": 18926.868255884016,
+      "required_pages_per_second_upper_estimate": 0.0043783978694942926
     },
     {
       "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": 28336.727157407488,
-      "required_pages_per_second_upper_estimate": 0.0017255580125686459
+      "required_token_positions_per_second": 28390.302383826027,
+      "required_pages_per_second_upper_estimate": 0.0017288204627700454
     },
     {
       "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": 28336.727157407488,
-      "required_pages_per_second_upper_estimate": 0.0021814396976584526
+      "required_token_positions_per_second": 28390.302383826027,
+      "required_pages_per_second_upper_estimate": 0.0021855640668938707
     },
     {
       "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": 28336.727157407488,
-      "required_pages_per_second_upper_estimate": 0.35184731436988576
+      "required_token_positions_per_second": 28390.302383826027,
+      "required_pages_per_second_upper_estimate": 0.3525125393772555
     }
   ],
   "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": 0.0977382867733354,
-    "producer_p95_seconds": 0.31534183399708127,
-    "gpu_update_seconds": 0.9756948939945788,
-    "producer_batches_per_second": 10.231405041087914,
-    "gpu_batches_per_second": 1.024910559802065,
-    "producer_headroom": 9.98272965697987,
+    "producer_batch_seconds": 0.12902346107793505,
+    "producer_p95_seconds": 0.30511665900121443,
+    "gpu_update_seconds": 0.9738536640508304,
+    "producer_batches_per_second": 7.7505284050314085,
+    "gpu_batches_per_second": 1.0268483211742632,
+    "producer_headroom": 7.547880485569876,
     "sustainable": true,
-    "buffer_seconds": 60.49308342766388,
+    "buffer_seconds": 60.37892717115149,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338248.51257743384,
-    "upload_bytes_per_second": 6384561.079798358,
+    "generation_bytes_per_second": 336095.03848775785,
+    "upload_bytes_per_second": 6375620.043960653,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.5802567767359,
+    "estimated_full_shards_per_hour": 72.11817136740257,
     "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": 1.0586995369994838,
-      "bytes": 7742624,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.8267605399996683,
       "bytes": 1098322,
@@ -562,6 +556,12 @@
       "bytes": 11363880,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 0.9065282010014926,
+      "bytes": 6177542,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -191,3 +191,4 @@
 {"step": 59100, "loss": 0.677880771458149, "lr": 0.0003, "grad_norm": 0.12769855087353754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8649907088, "tokens_per_second": 149227.430141975, "data_wait_s": 1.1377000191714615e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 59200, "loss": 0.6505282036960125, "lr": 0.0003, "grad_norm": 0.14494032713990299, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8664593793, "tokens_per_second": 149602.69704343355, "data_wait_s": 1.0273000953020528e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 59300, "loss": 0.6145510338246822, "lr": 0.0003, "grad_norm": 0.13085200480661205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8679171065, "tokens_per_second": 148698.6425819411, "data_wait_s": 1.0995001503033563e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 59400, "loss": 0.6136912349611521, "lr": 0.0003, "grad_norm": 0.12284896022984469, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8693908288, "tokens_per_second": 149322.3416025906, "data_wait_s": 8.126000466290861e-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 sha256:695bc8f23d7d889f76aee79ff19657fdfd1ea45a80c890d24a0501a0c6ea10b0
-size 411288
+oid sha256:d12d6ca7f6dcc38ea39094b70dfa79e0386ea4e676b81f7bb01e8616ce0e1420
+size 413406