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": 75104.31398698316,
-      "required_pages_per_second_upper_estimate": 0.006509851586248166
+      "required_token_positions_per_second": 75288.0456508552,
+      "required_pages_per_second_upper_estimate": 0.00652577698120895
     },
     {
       "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": 9388.039248372896,
-      "required_pages_per_second_upper_estimate": 0.0037450292198711087
+      "required_token_positions_per_second": 9411.0057063569,
+      "required_pages_per_second_upper_estimate": 0.0037541908833400753
     },
     {
       "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": 18776.07849674579,
-      "required_pages_per_second_upper_estimate": 0.004343515312526787
+      "required_token_positions_per_second": 18822.0114127138,
+      "required_pages_per_second_upper_estimate": 0.0043541410842442745
     },
     {
       "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": 28164.11774511869,
-      "required_pages_per_second_upper_estimate": 0.0017150470049718549
+      "required_token_positions_per_second": 28233.0171190707,
+      "required_pages_per_second_upper_estimate": 0.0017192426153584545
     },
     {
       "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": 28164.11774511869,
-      "required_pages_per_second_upper_estimate": 0.002168151747287026
+      "required_token_positions_per_second": 28233.0171190707,
+      "required_pages_per_second_upper_estimate": 0.0021734558118195265
     },
     {
       "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": 28164.11774511869,
-      "required_pages_per_second_upper_estimate": 0.3497040831558003
+      "required_token_positions_per_second": 28233.0171190707,
+      "required_pages_per_second_upper_estimate": 0.3505595827889132
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.27980502388308537,
-    "producer_p95_seconds": 0.3092805939995742,
-    "gpu_update_seconds": 0.9816746347324109,
-    "producer_batches_per_second": 3.5739172446662115,
-    "gpu_batches_per_second": 1.0186674531654618,
-    "producer_headroom": 3.5084239057215676,
+    "producer_batch_seconds": 0.06557538178105915,
+    "producer_p95_seconds": 0.22031322199836723,
+    "gpu_update_seconds": 0.9792789726792771,
+    "producer_batches_per_second": 15.249625283750019,
+    "gpu_batches_per_second": 1.0211594733460179,
+    "producer_headroom": 14.933637381614648,
     "sustainable": true,
-    "buffer_seconds": 60.86382735340948,
+    "buffer_seconds": 60.71529630611518,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347200.6479436076,
-    "upload_bytes_per_second": 5854763.545365107,
+    "generation_bytes_per_second": 384676.3892875459,
+    "upload_bytes_per_second": 5899698.42603734,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.50117663127108,
+    "estimated_full_shards_per_hour": 82.5425983330706,
     "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": 0.8353424750002887,
-      "bytes": 975468,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.1824829600009252,
       "bytes": 12495213,
@@ -562,6 +556,12 @@
       "bytes": 7464066,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.004503175001446,
+      "bytes": 5154959,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -169,3 +169,4 @@
 {"step": 331900, "loss": 0.5924331322312355, "lr": 0.0003, "grad_norm": 0.16922174544390134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48581636135, "tokens_per_second": 148586.97467842538, "data_wait_s": 1.1873002222273499e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 332000, "loss": 0.6186466738581657, "lr": 0.0003, "grad_norm": 0.14102725838681554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48596308602, "tokens_per_second": 149061.39464023084, "data_wait_s": 1.2320000678300858e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 332100, "loss": 0.59081706777215, "lr": 0.0003, "grad_norm": 0.1537119797637891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48610795305, "tokens_per_second": 147218.50037649143, "data_wait_s": 1.6715002857381478e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 332200, "loss": 0.6188142783939838, "lr": 0.0003, "grad_norm": 0.14477231213467215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48625459392, "tokens_per_second": 148293.71478576114, "data_wait_s": 1.2276999768801033e-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 sha256:25c41d6f52a44b20f3bc11f5ee22340fbbcdb3add953f7886f0d77b097a00cd0
-size 364692
+oid sha256:1f4cf1310ea0963b2bd8241915866eaff7c8ed20bb12d3f706f7eb0a903c78ff
+size 366810