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": 75255.16978992424,
-      "required_pages_per_second_upper_estimate": 0.006522927384906562
+      "required_token_positions_per_second": 76433.57027150605,
+      "required_pages_per_second_upper_estimate": 0.006625068152021357
     },
     {
       "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": 9406.89622374053,
-      "required_pages_per_second_upper_estimate": 0.0037525515492821647
+      "required_token_positions_per_second": 9554.196283938256,
+      "required_pages_per_second_upper_estimate": 0.0038113117456271962
     },
     {
       "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": 18813.79244748106,
-      "required_pages_per_second_upper_estimate": 0.004352239771286528
+      "required_token_positions_per_second": 19108.39256787651,
+      "required_pages_per_second_upper_estimate": 0.004420390324354978
     },
     {
       "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": 28220.688671221593,
-      "required_pages_per_second_upper_estimate": 0.0017184918775668108
+      "required_token_positions_per_second": 28662.588851814766,
+      "required_pages_per_second_upper_estimate": 0.0017454012800938684
     },
     {
       "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": 28220.688671221593,
-      "required_pages_per_second_upper_estimate": 0.002172506733776774
+      "required_token_positions_per_second": 28662.588851814766,
+      "required_pages_per_second_upper_estimate": 0.0022065254329368283
     },
     {
       "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": 28220.688671221593,
-      "required_pages_per_second_upper_estimate": 0.350406504727288
+      "required_token_positions_per_second": 28662.588851814766,
+      "required_pages_per_second_upper_estimate": 0.35589342602548846
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08387060671881841,
-    "producer_p95_seconds": 0.29111760900195804,
-    "gpu_update_seconds": 0.9797067790267784,
-    "producer_batches_per_second": 11.923128246258718,
-    "gpu_batches_per_second": 1.020713565944068,
-    "producer_headroom": 11.68116957006533,
+    "producer_batch_seconds": 0.06678454436729453,
+    "producer_p95_seconds": 0.22488066100049764,
+    "gpu_update_seconds": 0.9646023303386803,
+    "producer_batches_per_second": 14.973524330723983,
+    "gpu_batches_per_second": 1.0366966453926059,
+    "producer_headroom": 14.443496462799283,
     "sustainable": true,
-    "buffer_seconds": 60.74182029966026,
+    "buffer_seconds": 60.76994681133686,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334231.3324510622,
-    "upload_bytes_per_second": 3975982.2829143037,
+    "generation_bytes_per_second": 334190.96328930656,
+    "upload_bytes_per_second": 3969449.7594727287,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.71826343678379,
+    "estimated_full_shards_per_hour": 71.70960115441702,
     "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": 0.7973876159994688,
-      "bytes": 1188820,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.1932972649992735,
       "bytes": 7607279,
@@ -562,6 +556,12 @@
       "bytes": 6075989,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.8424355740025931,
+      "bytes": 669079,
+      "pages": 20,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -138,3 +138,4 @@
 {"step": 294000, "loss": 0.5516666574403644, "lr": 0.0003, "grad_norm": 0.14728797657168682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43033494776, "tokens_per_second": 149271.74959701372, "data_wait_s": 1.4050998288439587e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 294100, "loss": 0.5556954760104418, "lr": 0.0003, "grad_norm": 0.15022196021235193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43048083582, "tokens_per_second": 149508.01861430283, "data_wait_s": 1.106999843614176e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 294200, "loss": 0.633145309984684, "lr": 0.0003, "grad_norm": 0.153817176875024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43062718440, "tokens_per_second": 149666.37332655812, "data_wait_s": 8.15800012787804e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 294300, "loss": 0.5643345396965742, "lr": 0.0003, "grad_norm": 0.148157780776155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43077410234, "tokens_per_second": 139069.29664660958, "data_wait_s": 1.1342999641783535e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d95b8f6541305e1d03caa28c0fdc25e147ccfb1ac67f327d99dc65e792eee5dd
-size 298251
+oid sha256:6c6b3714ae118f316ecada7c5e3ac470cf12b31575dfc997f24cbd7f713be041
+size 300369