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": 77414.03639097755,
-      "required_pages_per_second_upper_estimate": 0.005355682662839792
+      "required_token_positions_per_second": 77151.83057433482,
+      "required_pages_per_second_upper_estimate": 0.005337542656043132
     },
     {
       "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": 9676.754548872193,
-      "required_pages_per_second_upper_estimate": 0.00405145505599925
+      "required_token_positions_per_second": 9643.978821791852,
+      "required_pages_per_second_upper_estimate": 0.004037732543505723
     },
     {
       "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": 19353.509097744387,
-      "required_pages_per_second_upper_estimate": 0.004030825558703252
+      "required_token_positions_per_second": 19287.957643583704,
+      "required_pages_per_second_upper_estimate": 0.004017172919509679
     },
     {
       "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": 29030.26364661658,
-      "required_pages_per_second_upper_estimate": 0.0017403199789686715
+      "required_token_positions_per_second": 28931.93646537556,
+      "required_pages_per_second_upper_estimate": 0.001734425414590185
     },
     {
       "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": 29030.26364661658,
-      "required_pages_per_second_upper_estimate": 0.0022569753078894197
+      "required_token_positions_per_second": 28931.93646537556,
+      "required_pages_per_second_upper_estimate": 0.002249330802043493
     },
     {
       "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": 29030.26364661658,
-      "required_pages_per_second_upper_estimate": 0.37236905179021024
+      "required_token_positions_per_second": 28931.93646537556,
+      "required_pages_per_second_upper_estimate": 0.3711078162847521
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11086969740637187,
-    "producer_p95_seconds": 0.3419581739944988,
-    "gpu_update_seconds": 0.9523854256563846,
-    "producer_batches_per_second": 9.019597089136894,
-    "gpu_batches_per_second": 1.049995068237,
-    "producer_headroom": 8.59013281298673,
+    "producer_batch_seconds": 0.1538938556329299,
+    "producer_p95_seconds": 0.327408754004864,
+    "gpu_update_seconds": 0.9556221732025398,
+    "producer_batches_per_second": 6.497985224213344,
+    "gpu_batches_per_second": 1.0464386742395673,
+    "producer_headroom": 6.209618761400749,
     "sustainable": true,
-    "buffer_seconds": 60.00028181635223,
+    "buffer_seconds": 60.20419691176001,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460566.9478545573,
-    "upload_bytes_per_second": 5035099.887963365,
+    "generation_bytes_per_second": 456430.08999708324,
+    "upload_bytes_per_second": 4982481.0933961775,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.82694555976428,
+    "estimated_full_shards_per_hour": 97.93927216467259,
     "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": 1.0368349029995443,
-      "bytes": 5950136,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 0.8421170019973943,
       "bytes": 2576016,
@@ -562,6 +556,12 @@
       "bytes": 10862357,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.9964048779947916,
+      "bytes": 1334873,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -375,3 +375,4 @@
 {"step": 477700, "loss": 0.546310305595398, "lr": 0.0003, "grad_norm": 0.14168281096870103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69919767493, "tokens_per_second": 151192.00415877238, "data_wait_s": 1.0787000064738095e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 477800, "loss": 0.600246787071228, "lr": 0.0003, "grad_norm": 0.16945478328948843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69934527837, "tokens_per_second": 153001.0171236244, "data_wait_s": 1.0626004950609058e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 477900, "loss": 0.587820291519165, "lr": 0.0003, "grad_norm": 0.14670491725173437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69949267109, "tokens_per_second": 153364.14544739126, "data_wait_s": 9.269999281968921e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 478000, "loss": 0.6480163931846619, "lr": 0.0003, "grad_norm": 0.21228140985040997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69964242413, "tokens_per_second": 155628.23587454614, "data_wait_s": 1.1324002116452903e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0d2fc04d617c9b76c46441182d89aac82f22c16860339128e8b10755c5399b1
-size 804132
+oid sha256:5baae70ad753433af1a0dac94ae11cc01f18bdb9dc4a728b36b6f54d6cd57a46
+size 806250