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": 76307.8151554823,
-      "required_pages_per_second_upper_estimate": 0.00527915171098133
+      "required_token_positions_per_second": 77265.04955957514,
+      "required_pages_per_second_upper_estimate": 0.0053453754081462974
     },
     {
       "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": 9538.476894435287,
-      "required_pages_per_second_upper_estimate": 0.003993561089652298
+      "required_token_positions_per_second": 9658.131194946893,
+      "required_pages_per_second_upper_estimate": 0.00404365784661058
     },
     {
       "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": 19076.953788870574,
-      "required_pages_per_second_upper_estimate": 0.003973226381118782
+      "required_token_positions_per_second": 19316.262389893785,
+      "required_pages_per_second_upper_estimate": 0.004023068051719712
     },
     {
       "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": 28615.430683305858,
-      "required_pages_per_second_upper_estimate": 0.001715451376231453
+      "required_token_positions_per_second": 28974.393584840676,
+      "required_pages_per_second_upper_estimate": 0.001736970654073851
     },
     {
       "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": 28615.430683305858,
-      "required_pages_per_second_upper_estimate": 0.0022247238696493812
+      "required_token_positions_per_second": 28974.393584840676,
+      "required_pages_per_second_upper_estimate": 0.002252631656332774
     },
     {
       "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": 28615.430683305858,
-      "required_pages_per_second_upper_estimate": 0.3670480199497936
+      "required_token_positions_per_second": 28974.393584840676,
+      "required_pages_per_second_upper_estimate": 0.37165241062634746
     }
   ],
   "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": 0.4206013964296176,
-    "producer_p95_seconds": 0.30261861799954204,
-    "gpu_update_seconds": 0.9661919929141498,
-    "producer_batches_per_second": 2.3775479788910725,
-    "gpu_batches_per_second": 1.0349909824691066,
-    "producer_headroom": 2.297167819973774,
+    "producer_batch_seconds": 0.425864461117186,
+    "producer_p95_seconds": 0.2988447570023709,
+    "gpu_update_seconds": 0.9542218690114487,
+    "producer_batches_per_second": 2.3481649475437867,
+    "gpu_batches_per_second": 1.0479743050072583,
+    "producer_headroom": 2.2406703449924024,
     "sustainable": true,
-    "buffer_seconds": 60.87009555359144,
+    "buffer_seconds": 60.11597774772127,
     "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": 328849.67455126916,
-    "upload_bytes_per_second": 5145435.921770859,
+    "generation_bytes_per_second": 329338.97385932686,
+    "upload_bytes_per_second": 5171083.961324681,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.56348493007236,
+    "estimated_full_shards_per_hour": 70.6684771712766,
     "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": 1.338383825001074,
-      "bytes": 7489532,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.0948180369996408,
       "bytes": 10021223,
@@ -562,6 +556,12 @@
       "bytes": 894969,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.5763037539982179,
+      "bytes": 10877110,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -276,3 +276,4 @@
 {"step": 952600, "loss": 0.5340831279754639, "lr": 3e-05, "grad_norm": 0.3493283224978132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139434654639, "tokens_per_second": 151007.9218448461, "data_wait_s": 1.3346001651370898e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952700, "loss": 0.5287927985191345, "lr": 3e-05, "grad_norm": 0.3457655496591882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139449234144, "tokens_per_second": 151632.50120178578, "data_wait_s": 9.719999070512131e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 952800, "loss": 0.5311126112937927, "lr": 3e-05, "grad_norm": 0.34549015143665845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139463934809, "tokens_per_second": 152711.52280609563, "data_wait_s": 1.7214002582477406e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 952900, "loss": 0.5499350428581238, "lr": 3e-05, "grad_norm": 0.354671723508823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139478697578, "tokens_per_second": 153814.60586857508, "data_wait_s": 1.2526001228252426e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ef98d6c56346d27608f2e2eeabab1de72ba5cef59ca5ee85a651501f41eb392
-size 592884
+oid sha256:b868ce802fd34f5dccbb8b6a208c9069ff8382edc7b16abed07a8ae58fd8d50d
+size 595002