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": 75837.24664270987,
-      "required_pages_per_second_upper_estimate": 0.006573380331245724
+      "required_token_positions_per_second": 75786.92808208054,
+      "required_pages_per_second_upper_estimate": 0.0065690188459416
     },
     {
       "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": 9479.655830338734,
-      "required_pages_per_second_upper_estimate": 0.0037815764442072496
+      "required_token_positions_per_second": 9473.366010260068,
+      "required_pages_per_second_upper_estimate": 0.0037790673409366796
     },
     {
       "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": 18959.311660677467,
-      "required_pages_per_second_upper_estimate": 0.004385903080208834
+      "required_token_positions_per_second": 18946.732020520136,
+      "required_pages_per_second_upper_estimate": 0.0043829930018525415
     },
     {
       "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": 28438.9674910162,
-      "required_pages_per_second_upper_estimate": 0.0017317839124718422
+      "required_token_positions_per_second": 28420.098030780202,
+      "required_pages_per_second_upper_estimate": 0.0017306348613438766
     },
     {
       "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": 28438.9674910162,
-      "required_pages_per_second_upper_estimate": 0.002189310441559009
+      "required_token_positions_per_second": 28420.098030780202,
+      "required_pages_per_second_upper_estimate": 0.0021878578182760325
     },
     {
       "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": 28438.9674910162,
-      "required_pages_per_second_upper_estimate": 0.3531167971369209
+      "required_token_positions_per_second": 28420.098030780202,
+      "required_pages_per_second_upper_estimate": 0.35288250159281076
     }
   ],
   "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": 0.4952689369141581,
-    "producer_p95_seconds": 0.40022729599877493,
-    "gpu_update_seconds": 0.9721871938119391,
-    "producer_batches_per_second": 2.019105026514764,
-    "gpu_batches_per_second": 1.0286084885350189,
-    "producer_headroom": 1.962948049738969,
+    "producer_batch_seconds": 0.13808338822678934,
+    "producer_p95_seconds": 0.3935333889967296,
+    "gpu_update_seconds": 0.9728326753150539,
+    "producer_batches_per_second": 7.242000742026923,
+    "gpu_batches_per_second": 1.0279259993771774,
+    "producer_headroom": 7.045254956499656,
     "sustainable": true,
-    "buffer_seconds": 60.27560601634022,
+    "buffer_seconds": 60.31562586953334,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 259670.11814768444,
-    "upload_bytes_per_second": 5451539.34341896,
+    "generation_bytes_per_second": 260471.58679771362,
+    "upload_bytes_per_second": 5515069.122818069,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.7191625435152,
+    "estimated_full_shards_per_hour": 55.89113905857617,
     "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": 1.6704007100015588,
-      "bytes": 10030549,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 2.951283398000669,
       "bytes": 2617533,
@@ -562,6 +556,12 @@
       "bytes": 1010925,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 0.9923871680002776,
+      "bytes": 11139165,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -363,3 +363,4 @@
 {"step": 151300, "loss": 0.5892051421105862, "lr": 0.0003, "grad_norm": 0.13786515756407766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22142229831, "tokens_per_second": 150042.62136134115, "data_wait_s": 9.256000339519233e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 151400, "loss": 0.5989663805812597, "lr": 0.0003, "grad_norm": 0.1467814174992331, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22156842202, "tokens_per_second": 137872.9737155881, "data_wait_s": 1.0670999472495168e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 151500, "loss": 0.6424321159720421, "lr": 0.0003, "grad_norm": 0.15058760000228508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22171636942, "tokens_per_second": 151178.1179345895, "data_wait_s": 2.3063003027345985e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 151600, "loss": 0.5758757181465626, "lr": 0.0003, "grad_norm": 0.16094720975034807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22186405283, "tokens_per_second": 151403.1334321257, "data_wait_s": 9.796000085771084e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c58bfb56143da2ebe4e3874474e6cdb421704ffa49e8c0f36999e6e02e405bd
-size 778716
+oid sha256:e8b073e6d22a084b957605ba7f11f5f88d5ad487b1280ad5f0115bd23cbc635d
+size 780834