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": 75670.77792034757,
-      "required_pages_per_second_upper_estimate": 0.006558951244302479
+      "required_token_positions_per_second": 74570.42690527807,
+      "required_pages_per_second_upper_estimate": 0.006463575607130407
     },
     {
       "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": 9458.847240043446,
-      "required_pages_per_second_upper_estimate": 0.0037732755864223096
+      "required_token_positions_per_second": 9321.30336315976,
+      "required_pages_per_second_upper_estimate": 0.0037184072774691873
     },
     {
       "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": 18917.69448008689,
-      "required_pages_per_second_upper_estimate": 0.004376275677916567
+      "required_token_positions_per_second": 18642.60672631952,
+      "required_pages_per_second_upper_estimate": 0.004312638967461598
     },
     {
       "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": 28376.541720130335,
-      "required_pages_per_second_upper_estimate": 0.0017279825105476006
+      "required_token_positions_per_second": 27963.910089479275,
+      "required_pages_per_second_upper_estimate": 0.001702855409151804
     },
     {
       "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": 28376.541720130335,
-      "required_pages_per_second_upper_estimate": 0.002184504732910621
+      "required_token_positions_per_second": 27963.910089479275,
+      "required_pages_per_second_upper_estimate": 0.0021527392077456426
     },
     {
       "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": 28376.541720130335,
-      "required_pages_per_second_upper_estimate": 0.3523416779881338
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27963.910089479275,
+      "required_pages_per_second_upper_estimate": 0.34721817412467904
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.1000281149296427,
-    "producer_p95_seconds": 0.19783594399996218,
-    "gpu_update_seconds": 0.9743259158457104,
-    "producer_batches_per_second": 9.997189297262826,
-    "gpu_batches_per_second": 1.0263506119838808,
-    "producer_headroom": 9.740520617938538,
+    "producer_batch_seconds": 0.209447565890585,
+    "producer_p95_seconds": 0.20964279399959196,
+    "gpu_update_seconds": 0.9887029357315046,
+    "producer_batches_per_second": 4.7744646529929025,
+    "gpu_batches_per_second": 1.011426146176189,
+    "producer_headroom": 4.720527218960382,
     "sustainable": true,
-    "buffer_seconds": 60.40820678243404,
+    "buffer_seconds": 60.31087907962178,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317338.72879478324,
-    "upload_bytes_per_second": 3911356.0008797403,
+    "generation_bytes_per_second": 314610.50391598814,
+    "upload_bytes_per_second": 3960470.188988808,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.09350393183348,
+    "estimated_full_shards_per_hour": 67.50809038266881,
     "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": 1.0635986559996127,
-      "bytes": 6186061,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1765199270002995,
       "bytes": 6893561,
@@ -562,6 +556,12 @@
       "bytes": 2401103,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.182281362998765,
+      "bytes": 11896636,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -91,3 +91,4 @@
 {"step": 324100, "loss": 0.5908871721476316, "lr": 0.0003, "grad_norm": 0.17862144998034735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47439485499, "tokens_per_second": 150673.0591248944, "data_wait_s": 2.031700023508165e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 324200, "loss": 0.610282639041543, "lr": 0.0003, "grad_norm": 0.1444129732046015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47453953101, "tokens_per_second": 147405.01491035265, "data_wait_s": 1.0828000085894018e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 324300, "loss": 0.6009922064840794, "lr": 0.0003, "grad_norm": 0.1589253453260637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47468469558, "tokens_per_second": 147243.66094686935, "data_wait_s": 1.6060999769251794e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 324400, "loss": 0.5966446306556463, "lr": 0.0003, "grad_norm": 0.19220411014517647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47482978852, "tokens_per_second": 147404.66821660692, "data_wait_s": 1.2679000064963475e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a233ce0d12ad1f4c0ea60b6557b813579beab0d270a2f96ffdf3dbe51e12be3
-size 197922
+oid sha256:fb78c31d8e7c9c5792afb46e25dd4de3e481ef5ac297dff03a85ec2f8dc119e7
+size 200040