Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76961.14952444332,
-      "required_pages_per_second_upper_estimate": 0.005324350898570646
+      "required_token_positions_per_second": 76456.7235779744,
+      "required_pages_per_second_upper_estimate": 0.0052894535411644765
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9620.143690555415,
-      "required_pages_per_second_upper_estimate": 0.004027753271791165
+      "required_token_positions_per_second": 9557.0904472468,
+      "required_pages_per_second_upper_estimate": 0.004001354195519296
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19240.28738111083,
-      "required_pages_per_second_upper_estimate": 0.0040072444609859405
+      "required_token_positions_per_second": 19114.1808944936,
+      "required_pages_per_second_upper_estimate": 0.003980979805475264
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28860.431071666244,
-      "required_pages_per_second_upper_estimate": 0.0017301387754197258
+      "required_token_positions_per_second": 28671.271341740397,
+      "required_pages_per_second_upper_estimate": 0.001718798938440855
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28860.431071666244,
-      "required_pages_per_second_upper_estimate": 0.0022437715722015816
+      "required_token_positions_per_second": 28671.271341740397,
+      "required_pages_per_second_upper_estimate": 0.002229065235225568
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28860.431071666244,
-      "required_pages_per_second_upper_estimate": 0.370190621870759
+      "required_token_positions_per_second": 28671.271341740397,
+      "required_pages_per_second_upper_estimate": 0.36776428395916416
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08196175659371363,
-    "producer_p95_seconds": 0.23701396199976443,
-    "gpu_update_seconds": 0.9579898488468335,
-    "producer_batches_per_second": 12.200812202659634,
-    "gpu_batches_per_second": 1.043852396978669,
-    "producer_headroom": 11.688254237834505,
+    "producer_batch_seconds": 0.11916180506248963,
+    "producer_p95_seconds": 0.2299430940001912,
+    "gpu_update_seconds": 0.9643102208638132,
+    "producer_batches_per_second": 8.391950755324578,
+    "gpu_batches_per_second": 1.0370106822099392,
+    "producer_headroom": 8.092443886345288,
     "sustainable": true,
-    "buffer_seconds": 60.353360477350506,
+    "buffer_seconds": 60.75154391442023,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 172,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 464154.6877819868,
-    "upload_bytes_per_second": 4011505.656124086,
+    "generation_bytes_per_second": 588689.8765327701,
+    "upload_bytes_per_second": 3947914.7908177157,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.5967910298796,
+    "estimated_full_shards_per_hour": 63.15957175248779,
     "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": 1.649759572000221,
-      "bytes": 7616382,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.8874110320002728,
       "bytes": 11916705,
@@ -562,6 +556,12 @@
       "bytes": 6129393,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.743864690999544,
+      "bytes": 1054003,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -131,3 +131,4 @@
 {"step": 813100, "loss": 0.5329916477203369, "lr": 3e-05, "grad_norm": 0.291605878109602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119014428690, "tokens_per_second": 150627.43206019083, "data_wait_s": 1.3560000297729857e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 813200, "loss": 0.5572542548179626, "lr": 3e-05, "grad_norm": 0.3302031908063002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119029057222, "tokens_per_second": 151758.50868091194, "data_wait_s": 1.0109000868396834e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 813300, "loss": 0.5391030311584473, "lr": 3e-05, "grad_norm": 0.2895030831096297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119043829996, "tokens_per_second": 152591.81621550187, "data_wait_s": 7.164999260567129e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 813400, "loss": 0.5715728402137756, "lr": 3e-05, "grad_norm": 0.29769786922742275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119058540394, "tokens_per_second": 152459.69318877018, "data_wait_s": 7.018999895080924e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:67275e2f16f62bd70aabf80628ed6a5e8a219e10a207b8b62a30eb272095bbf7
-size 283425
+oid sha256:ac048a851375d1b97d8b56f8261615b869de29c42402bbd5f70f71cc8b61fc24
+size 285543