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": 75650.76339983131,
-      "required_pages_per_second_upper_estimate": 0.006557216436390468
+      "required_token_positions_per_second": 74776.88341817356,
+      "required_pages_per_second_upper_estimate": 0.0064814707344089395
     },
     {
       "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": 9456.345424978914,
-      "required_pages_per_second_upper_estimate": 0.0037722775749875993
+      "required_token_positions_per_second": 9347.110427271695,
+      "required_pages_per_second_upper_estimate": 0.0037287021011934316
     },
     {
       "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": 18912.690849957828,
-      "required_pages_per_second_upper_estimate": 0.004375118176147092
+      "required_token_positions_per_second": 18694.22085454339,
+      "required_pages_per_second_upper_estimate": 0.004324578987648559
     },
     {
       "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": 28369.036274936745,
-      "required_pages_per_second_upper_estimate": 0.0017275254683133386
+      "required_token_positions_per_second": 28041.331281815084,
+      "required_pages_per_second_upper_estimate": 0.0017075699535674513
     },
     {
       "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": 28369.036274936745,
-      "required_pages_per_second_upper_estimate": 0.0021839269429632163
+      "required_token_positions_per_second": 28041.331281815084,
+      "required_pages_per_second_upper_estimate": 0.00215869930544723
     },
     {
       "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": 28369.036274936745,
-      "required_pages_per_second_upper_estimate": 0.35224848547793863
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28041.331281815084,
+      "required_pages_per_second_upper_estimate": 0.34817948622142725
     }
   ],
   "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": 0.09811553094542091,
-    "producer_p95_seconds": 0.34001389499826473,
-    "gpu_update_seconds": 0.9745836880763638,
-    "producer_batches_per_second": 10.192066336126477,
-    "gpu_batches_per_second": 1.0260791476756634,
-    "producer_headroom": 9.933021598981092,
+    "producer_batch_seconds": 0.4155778188281545,
+    "producer_p95_seconds": 0.3382514959994296,
+    "gpu_update_seconds": 0.9859731594815485,
+    "producer_batches_per_second": 2.406288196082741,
+    "gpu_batches_per_second": 1.0142263918480572,
+    "producer_headroom": 2.3725355753148563,
     "sustainable": true,
-    "buffer_seconds": 60.42418866073455,
+    "buffer_seconds": 60.14436272837446,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384380.5326686332,
-    "upload_bytes_per_second": 3939907.144593242,
+    "generation_bytes_per_second": 341109.2099537289,
+    "upload_bytes_per_second": 4007982.342240348,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.47911438984153,
+    "estimated_full_shards_per_hour": 73.19409584006215,
     "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": 1.065639946002193,
-      "bytes": 5791258,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.1834909390017856,
       "bytes": 7846092,
@@ -562,6 +556,12 @@
       "bytes": 5901703,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1044195929971465,
+      "bytes": 13283238,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -237,3 +237,4 @@
 {"step": 138700, "loss": 0.5369933377951384, "lr": 0.0003, "grad_norm": 0.13935597412628523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20298343512, "tokens_per_second": 148685.80107937855, "data_wait_s": 9.620001947041601e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 138800, "loss": 0.5648066718131304, "lr": 0.0003, "grad_norm": 0.15375385803934544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20313041480, "tokens_per_second": 150122.57332411828, "data_wait_s": 1.0962998203467578e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 138900, "loss": 0.5808432884514332, "lr": 0.0003, "grad_norm": 0.15437790619225644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20327720677, "tokens_per_second": 149687.26270518242, "data_wait_s": 1.1454001651145518e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 139000, "loss": 0.589487548917532, "lr": 0.0003, "grad_norm": 0.1428047880376676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20342441712, "tokens_per_second": 150873.92026288502, "data_wait_s": 1.2471999070839956e-05, "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 sha256:3f7a52080d9d0266fcdacb43a1e31e3d8efb4314cf19da53d5b00aea42f17b91
-size 509499
+oid sha256:abe6a636d4dd6aa6200f33fbe5d5c52db4b06fd86c1487463fe452363e9a28bd
+size 511617