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": 76721.60466180158,
-      "required_pages_per_second_upper_estimate": 0.005307778629152432
+      "required_token_positions_per_second": 76695.32894137189,
+      "required_pages_per_second_upper_estimate": 0.005305960813845035
     },
     {
       "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": 9590.200582725198,
-      "required_pages_per_second_upper_estimate": 0.0040152167178258485
+      "required_token_positions_per_second": 9586.916117671486,
+      "required_pages_per_second_upper_estimate": 0.004013841580895289
     },
     {
       "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": 19180.401165450396,
-      "required_pages_per_second_upper_estimate": 0.003994771741569583
+      "required_token_positions_per_second": 19173.832235342972,
+      "required_pages_per_second_upper_estimate": 0.003993403606662543
     },
     {
       "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": 28770.601748175595,
-      "required_pages_per_second_upper_estimate": 0.0017247536446378965
+      "required_token_positions_per_second": 28760.748353014456,
+      "required_pages_per_second_upper_estimate": 0.0017241629486432495
     },
     {
       "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": 28770.601748175595,
-      "required_pages_per_second_upper_estimate": 0.0022367877374176205
+      "required_token_positions_per_second": 28760.748353014456,
+      "required_pages_per_second_upper_estimate": 0.0022360216792843466
     },
     {
       "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": 28770.601748175595,
-      "required_pages_per_second_upper_estimate": 0.3690383877602339
+      "required_token_positions_per_second": 28760.748353014456,
+      "required_pages_per_second_upper_estimate": 0.3689119989868583
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13835116811736725,
-    "producer_p95_seconds": 0.32738750600401545,
-    "gpu_update_seconds": 0.9609809430472972,
-    "producer_batches_per_second": 7.227983786531325,
-    "gpu_batches_per_second": 1.040603361840842,
-    "producer_headroom": 6.945954675511446,
+    "producer_batch_seconds": 0.14030650369534214,
+    "producer_p95_seconds": 0.32547561899991706,
+    "gpu_update_seconds": 0.9613101738746019,
+    "producer_batches_per_second": 7.127253360766324,
+    "gpu_batches_per_second": 1.040246974573729,
+    "producer_headroom": 6.851501167486616,
     "sustainable": true,
-    "buffer_seconds": 60.541799411979724,
+    "buffer_seconds": 60.56254095409992,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317124.9127511041,
+    "generation_bytes_per_second": 318157.2893284005,
     "upload_bytes_per_second": 4688621.170155269,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.0476239862427,
+    "estimated_full_shards_per_hour": 68.26914796723376,
     "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": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -379,3 +379,4 @@
 {"step": 797900, "loss": 0.5144842863082886, "lr": 3e-05, "grad_norm": 0.27186701635641053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116789626837, "tokens_per_second": 153466.22160680825, "data_wait_s": 8.831004379317164e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 798000, "loss": 0.5630819797515869, "lr": 3e-05, "grad_norm": 0.2660672224736182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116804324569, "tokens_per_second": 152752.5646681396, "data_wait_s": 1.0050003766082227e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 798100, "loss": 0.44999396800994873, "lr": 3e-05, "grad_norm": 0.26055396723796714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116818909689, "tokens_per_second": 151903.89110008214, "data_wait_s": 1.846699888119474e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 798200, "loss": 0.5165267586708069, "lr": 3e-05, "grad_norm": 0.269538701766825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116833409310, "tokens_per_second": 151597.68100082804, "data_wait_s": 9.444003808312118e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:537e918b7d30e4b87411867165be300e3e06f996fbcc1eee6cc36b8fc6d5894c
-size 812604
+oid sha256:29476b63e28f7f0371a033440bc48239a70e2fac642db0234f636e30c47d8426
+size 814722