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": 76355.06690974647,
-      "required_pages_per_second_upper_estimate": 0.0052824206969805014
+      "required_token_positions_per_second": 76879.57661690105,
+      "required_pages_per_second_upper_estimate": 0.005318707495551649
     },
     {
       "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": 9544.383363718309,
-      "required_pages_per_second_upper_estimate": 0.003996034004999995
+      "required_token_positions_per_second": 9609.94707711263,
+      "required_pages_per_second_upper_estimate": 0.004023484162672173
     },
     {
       "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": 19088.766727436618,
-      "required_pages_per_second_upper_estimate": 0.00397568670469395
+      "required_token_positions_per_second": 19219.89415422526,
+      "required_pages_per_second_upper_estimate": 0.004002997089631235
     },
     {
       "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": 28633.150091154926,
-      "required_pages_per_second_upper_estimate": 0.0017165136276760355
+      "required_token_positions_per_second": 28829.84123133789,
+      "required_pages_per_second_upper_estimate": 0.0017283049612000227
     },
     {
       "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": 28633.150091154926,
-      "required_pages_per_second_upper_estimate": 0.002226101475670208
+      "required_token_positions_per_second": 28829.84123133789,
+      "required_pages_per_second_upper_estimate": 0.00224139334666654
     },
     {
       "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": 28633.150091154926,
-      "required_pages_per_second_upper_estimate": 0.3672753054880636
+      "required_token_positions_per_second": 28829.84123133789,
+      "required_pages_per_second_upper_estimate": 0.36979824824383845
     }
   ],
   "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": 0.1715052397890986,
-    "producer_p95_seconds": 0.293062724002084,
-    "gpu_update_seconds": 0.9655940723246078,
-    "producer_batches_per_second": 5.8307256456403795,
-    "gpu_batches_per_second": 1.0356318754034624,
-    "producer_headroom": 5.630114120781422,
+    "producer_batch_seconds": 0.10866544688275326,
+    "producer_p95_seconds": 0.2785734600001888,
+    "gpu_update_seconds": 0.9590063218921498,
+    "producer_batches_per_second": 9.202557286484726,
+    "gpu_batches_per_second": 1.0427459936103114,
+    "producer_headroom": 8.825310615313521,
     "sustainable": true,
-    "buffer_seconds": 60.832426556450294,
+    "buffer_seconds": 60.41739827920544,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440377.0369773188,
-    "upload_bytes_per_second": 5129436.425761475,
+    "generation_bytes_per_second": 423384.287767717,
+    "upload_bytes_per_second": 5198158.87594895,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.49466068257973,
+    "estimated_full_shards_per_hour": 90.84841227315552,
     "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": 0.88294980000137,
-      "bytes": 116127,
-      "pages": 4,
-      "shards": 1
-    },
     {
       "seconds": 1.2832525270005135,
       "bytes": 10171575,
@@ -562,6 +556,12 @@
       "bytes": 5474226,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.198957466000138,
+      "bytes": 7513050,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 627800, "loss": 0.5317293405532837, "lr": 0.0001, "grad_norm": 0.17683979314352088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91893147045, "tokens_per_second": 149404.02428802187, "data_wait_s": 1.501199949416332e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 627900, "loss": 0.528694748878479, "lr": 0.0001, "grad_norm": 0.15786228055553655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91907824827, "tokens_per_second": 151905.32613744913, "data_wait_s": 1.595100184204057e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 628000, "loss": 0.5480430126190186, "lr": 0.0001, "grad_norm": 0.1529228731650541, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91922483197, "tokens_per_second": 151008.74539826976, "data_wait_s": 1.895099921966903e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 628100, "loss": 0.5606343746185303, "lr": 0.0001, "grad_norm": 0.1497014292572719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91937212305, "tokens_per_second": 152742.78066113414, "data_wait_s": 1.5298999642254785e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:85ab7101b6e3741d49fc2df7ac9b9091476e41b84cd37c8d366f89efbd45d78e
-size 503928
+oid sha256:7811140fc453eea47678ce3cb56c9ff2e959735867deebc752ad6e45267261fc
+size 506046