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": 76156.03701224625,
-      "required_pages_per_second_upper_estimate": 0.006601012272510333
+      "required_token_positions_per_second": 74524.44476239788,
+      "required_pages_per_second_upper_estimate": 0.006459589991526227
     },
     {
       "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": 9519.504626530781,
-      "required_pages_per_second_upper_estimate": 0.003797472724800854
+      "required_token_positions_per_second": 9315.555595299735,
+      "required_pages_per_second_upper_estimate": 0.003716114406933036
     },
     {
       "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": 19039.009253061562,
-      "required_pages_per_second_upper_estimate": 0.004404339715577175
+      "required_token_positions_per_second": 18631.11119059947,
+      "required_pages_per_second_upper_estimate": 0.004309979677587284
     },
     {
       "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": 28558.51387959234,
-      "required_pages_per_second_upper_estimate": 0.001739063660324702
+      "required_token_positions_per_second": 27946.666785899204,
+      "required_pages_per_second_upper_estimate": 0.0017018053824323998
     },
     {
       "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": 28558.51387959234,
-      "required_pages_per_second_upper_estimate": 0.002198513453477184
+      "required_token_positions_per_second": 27946.666785899204,
+      "required_pages_per_second_upper_estimate": 0.0021514117705035292
     },
     {
       "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": 28558.51387959234,
-      "required_pages_per_second_upper_estimate": 0.35460116318701146
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27946.666785899204,
+      "required_pages_per_second_upper_estimate": 0.34700407000383926
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.14275012733592796,
-    "producer_p95_seconds": 0.14122689800001353,
-    "gpu_update_seconds": 0.9681176029176018,
-    "producer_batches_per_second": 7.005247691630715,
-    "gpu_batches_per_second": 1.0329323596496072,
-    "producer_headroom": 6.781903603065592,
+    "producer_batch_seconds": 0.1034238145078259,
+    "producer_p95_seconds": 0.14257186500003627,
+    "gpu_update_seconds": 0.9893129728784007,
+    "producer_batches_per_second": 9.668952984946536,
+    "gpu_batches_per_second": 1.0108024734483219,
+    "producer_headroom": 9.565620622158944,
     "sustainable": true,
-    "buffer_seconds": 60.023291380891315,
+    "buffer_seconds": 60.34809134558244,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 129,
+    "upload_pages_per_commit": 142,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 804610.1421376357,
-    "upload_bytes_per_second": 4608162.827356506,
+    "generation_bytes_per_second": 750496.0166078976,
+    "upload_bytes_per_second": 4417317.494560352,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.3253030686226,
+    "estimated_full_shards_per_hour": 80.51948725546691,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 10769730,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2535849739999776,
+      "bytes": 3640398,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -4,3 +4,4 @@
 {"step": 400, "loss": 1.7191495224833488, "lr": 0.00011999999999999999, "grad_norm": 2.6239859511966785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58978491, "tokens_per_second": 150971.93971749238, "data_wait_s": 1.4216999943528208e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 500, "loss": 1.528922200202942, "lr": 0.00015, "grad_norm": 1.9988208797091183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 73647586, "tokens_per_second": 150271.54409745906, "data_wait_s": 1.2140999842813471e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 600, "loss": 1.4354848563671112, "lr": 0.00017999999999999998, "grad_norm": 1.4426431949597145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 88325921, "tokens_per_second": 150330.61897931347, "data_wait_s": 9.991000069931033e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 700, "loss": 1.4169969446957111, "lr": 0.00020999999999999998, "grad_norm": 1.1299403160160901, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 103081101, "tokens_per_second": 149879.6582243547, "data_wait_s": 1.00040001598245e-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 sha256:43201dceca320b22042014c83dd4557d07a4941a18592a95e6c52312d99cfa09
-size 12624
+oid sha256:fa1e7302f1716beb9650cbc67b221e6cd286ee6c4f7978d2f6f6db3cb2fb2fd2
+size 14707