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": 75423.03437789231,
-      "required_pages_per_second_upper_estimate": 0.006537477461942725
+      "required_token_positions_per_second": 73307.36100447136,
+      "required_pages_per_second_upper_estimate": 0.0063540962560597394
     },
     {
       "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": 9427.879297236539,
-      "required_pages_per_second_upper_estimate": 0.0037609220110246286
+      "required_token_positions_per_second": 9163.42012555892,
+      "required_pages_per_second_upper_estimate": 0.0036554252934254505
     },
     {
       "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": 18855.758594473078,
-      "required_pages_per_second_upper_estimate": 0.0043619479008141695
+      "required_token_positions_per_second": 18326.84025111784,
+      "required_pages_per_second_upper_estimate": 0.004239591950723848
     },
     {
       "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": 28283.637891709615,
-      "required_pages_per_second_upper_estimate": 0.001722325154825496
+      "required_token_positions_per_second": 27490.26037667676,
+      "required_pages_per_second_upper_estimate": 0.0016740126266901194
     },
     {
       "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": 28283.637891709615,
-      "required_pages_per_second_upper_estimate": 0.0021773527390245373
+      "required_token_positions_per_second": 27490.26037667676,
+      "required_pages_per_second_upper_estimate": 0.0021162763417077826
     },
     {
       "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": 28283.637891709615,
-      "required_pages_per_second_upper_estimate": 0.3511881233682607
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27490.26037667676,
+      "required_pages_per_second_upper_estimate": 0.3413370298952874
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10480932768739137,
-    "producer_p95_seconds": 0.19093247900127608,
-    "gpu_update_seconds": 0.9775263035772378,
-    "producer_batches_per_second": 9.541135527389711,
-    "gpu_batches_per_second": 1.0229903751341731,
-    "producer_headroom": 9.326710944018725,
+    "producer_batch_seconds": 0.20125084082809508,
+    "producer_p95_seconds": 0.1800921499998367,
+    "gpu_update_seconds": 1.005738018525902,
+    "producer_batches_per_second": 4.968923339079027,
+    "gpu_batches_per_second": 0.9942947184851259,
+    "producer_headroom": 4.9974351132524495,
     "sustainable": true,
-    "buffer_seconds": 60.60663082178874,
+    "buffer_seconds": 60.34428111155412,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 59,
+    "upload_pages_per_commit": 63,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 426054.75137415825,
-    "upload_bytes_per_second": 5293868.982980686,
+    "generation_bytes_per_second": 387193.7202237535,
+    "upload_bytes_per_second": 4803572.700355249,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.42143159788667,
+    "estimated_full_shards_per_hour": 83.0827589515157,
     "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": [
@@ -214,6 +214,12 @@
       "bytes": 6094729,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2288260940003966,
+      "bytes": 2371624,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -6,3 +6,4 @@
 {"step": 280800, "loss": 0.6316648758947849, "lr": 0.0003, "grad_norm": 0.17616777584734625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41101531907, "tokens_per_second": 143727.64187390768, "data_wait_s": 9.895999028231017e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 280900, "loss": 0.5992213487625122, "lr": 0.0003, "grad_norm": 0.1497594529084744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41116106281, "tokens_per_second": 149885.96635611446, "data_wait_s": 1.329700171481818e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 281000, "loss": 0.6061381213366985, "lr": 0.0003, "grad_norm": 0.15370715606758695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41130749779, "tokens_per_second": 149028.8243692584, "data_wait_s": 1.2833999790018424e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 281100, "loss": 0.610587639734149, "lr": 0.0003, "grad_norm": 0.14617332899983632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41145514716, "tokens_per_second": 150904.03889516508, "data_wait_s": 9.845000022323802e-06, "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 sha256:9b8300e9e8f83f29ca48eca6ac2b4e1479701258b90f8937f32b17da2ac96563
-size 17109
+oid sha256:d65dafeb87a6a6818a44a52d2934b3421281e5c6e6cf0966375871ca45350326
+size 19227