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": 75601.86866191775,
-      "required_pages_per_second_upper_estimate": 0.006552978364430701
+      "required_token_positions_per_second": 74063.33551674566,
+      "required_pages_per_second_upper_estimate": 0.006419622210783791
     },
     {
       "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": 9450.233582739718,
-      "required_pages_per_second_upper_estimate": 0.0037698394697381994
+      "required_token_positions_per_second": 9257.916939593208,
+      "required_pages_per_second_upper_estimate": 0.003693121485397003
     },
     {
       "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": 18900.467165479437,
-      "required_pages_per_second_upper_estimate": 0.004372290441872514
+      "required_token_positions_per_second": 18515.833879186415,
+      "required_pages_per_second_upper_estimate": 0.004283312300403262
     },
     {
       "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": 28350.70074821916,
-      "required_pages_per_second_upper_estimate": 0.0017264089309353117
+      "required_token_positions_per_second": 27773.750818779623,
+      "required_pages_per_second_upper_estimate": 0.001691275707254783
     },
     {
       "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": 28350.70074821916,
-      "required_pages_per_second_upper_estimate": 0.00218251542336051
+      "required_token_positions_per_second": 27773.750818779623,
+      "required_pages_per_second_upper_estimate": 0.0021381002206926313
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28350.70074821916,
-      "required_pages_per_second_upper_estimate": 0.35202081960116666
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27773.750818779623,
+      "required_pages_per_second_upper_estimate": 0.34485703240472854
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1115278211328814,
-    "producer_p95_seconds": 0.3788880829997652,
-    "gpu_update_seconds": 0.9752139901422614,
-    "producer_batches_per_second": 8.966372604092532,
-    "gpu_batches_per_second": 1.025415970349362,
-    "producer_headroom": 8.744132004339338,
+    "producer_batch_seconds": 0.45545081864068493,
+    "producer_p95_seconds": 0.38295882899910794,
+    "gpu_update_seconds": 0.9954723141429427,
+    "producer_batches_per_second": 2.195626748425985,
+    "gpu_batches_per_second": 1.0045482790357212,
+    "producer_headroom": 2.1856856402497598,
     "sustainable": true,
-    "buffer_seconds": 60.463267388820206,
+    "buffer_seconds": 60.72381116271951,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390148.1906052726,
-    "upload_bytes_per_second": 6240443.286975443,
+    "generation_bytes_per_second": 353925.6583919374,
+    "upload_bytes_per_second": 6320002.38046379,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.7167195188392,
+    "estimated_full_shards_per_hour": 75.94420732325163,
     "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": 1.2853242279998085,
-      "bytes": 7445426,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.7597367620001023,
       "bytes": 831447,
@@ -562,6 +556,12 @@
       "bytes": 1148192,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.378931018000003,
+      "bytes": 13331524,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -312,3 +312,4 @@
 {"step": 31200, "loss": 0.664759548380971, "lr": 0.0003, "grad_norm": 0.13522052249851715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4565619590, "tokens_per_second": 149727.55217065001, "data_wait_s": 8.682000043336302e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 31300, "loss": 0.6205551885068417, "lr": 0.0003, "grad_norm": 0.12686401793089822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4580321298, "tokens_per_second": 148694.87510512723, "data_wait_s": 1.2466000043787062e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 31400, "loss": 0.6879486478865147, "lr": 0.0003, "grad_norm": 0.16777613929339316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4595072175, "tokens_per_second": 150432.5182586362, "data_wait_s": 1.0510000720387325e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 31500, "loss": 0.7267534174025059, "lr": 0.0003, "grad_norm": 0.13736877923239438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4609858831, "tokens_per_second": 149405.9264729884, "data_wait_s": 1.3778000720776618e-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:dc4e36d9a26038571493277cba2f40229681886b245149b0c553fb83b6e8c0a5
-size 664141
+oid sha256:e94967f4ee0c448075694471a7a1d35d1da50cfbe18246dde02eb925d9011ae4
+size 666259