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": 77229.77478248172,
-      "required_pages_per_second_upper_estimate": 0.005342935017218215
+      "required_token_positions_per_second": 76327.57162622061,
+      "required_pages_per_second_upper_estimate": 0.005280518509468334
     },
     {
       "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": 9653.721847810215,
-      "required_pages_per_second_upper_estimate": 0.004041811745042092
+      "required_token_positions_per_second": 9540.946453277576,
+      "required_pages_per_second_upper_estimate": 0.003994595042369312
     },
     {
       "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": 19307.44369562043,
-      "required_pages_per_second_upper_estimate": 0.00402123135026718
+      "required_token_positions_per_second": 19081.892906555153,
+      "required_pages_per_second_upper_estimate": 0.003974255069079187
     },
     {
       "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": 28961.165543430645,
-      "required_pages_per_second_upper_estimate": 0.0017361776531893693
+      "required_token_positions_per_second": 28622.839359832727,
+      "required_pages_per_second_upper_estimate": 0.0017158955150768447
     },
     {
       "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": 28961.165543430645,
-      "required_pages_per_second_upper_estimate": 0.002251603234297149
+      "required_token_positions_per_second": 28622.839359832727,
+      "required_pages_per_second_upper_estimate": 0.0022252998616620213
     },
     {
       "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": 28961.165543430645,
-      "required_pages_per_second_upper_estimate": 0.3714827355143039
+      "required_token_positions_per_second": 28622.839359832727,
+      "required_pages_per_second_upper_estimate": 0.3671430504974632
     }
   ],
   "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": 0.0574585945312549,
-    "producer_p95_seconds": 0.19607393300066178,
-    "gpu_update_seconds": 0.954657710807205,
-    "producer_batches_per_second": 17.403836765552015,
-    "gpu_batches_per_second": 1.0474958602224624,
-    "producer_headroom": 16.614706965864155,
+    "producer_batch_seconds": 0.09984876964844602,
+    "producer_p95_seconds": 0.2026637640010449,
+    "gpu_update_seconds": 0.9659419057774978,
+    "producer_batches_per_second": 10.015145940414332,
+    "gpu_batches_per_second": 1.0352589467532092,
+    "producer_headroom": 9.67404915632359,
     "sustainable": true,
-    "buffer_seconds": 60.14343578085391,
+    "buffer_seconds": 60.85434006398236,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301799.47713915625,
-    "upload_bytes_per_second": 5659223.606299439,
+    "generation_bytes_per_second": 313655.62182173267,
+    "upload_bytes_per_second": 5568236.237819763,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.75914226180092,
+    "estimated_full_shards_per_hour": 67.30319491375909,
     "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": 1.0907009550001021,
-      "bytes": 7267703,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.059233770999981,
       "bytes": 6001877,
@@ -562,6 +556,12 @@
       "bytes": 12132833,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.2424271460004093,
+      "bytes": 1081337,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -97,3 +97,4 @@
 {"step": 614100, "loss": 0.5311686396598816, "lr": 0.0001, "grad_norm": 0.15072206927000883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89886189019, "tokens_per_second": 150754.44136008003, "data_wait_s": 5.9630001487676054e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 614200, "loss": 0.514626681804657, "lr": 0.0001, "grad_norm": 0.15638066934972122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89900729754, "tokens_per_second": 151111.94238210047, "data_wait_s": 1.280899959965609e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 614300, "loss": 0.5313458442687988, "lr": 0.0001, "grad_norm": 0.1551929440788104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89915535292, "tokens_per_second": 154554.22675837556, "data_wait_s": 1.3519000276573934e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 614400, "loss": 0.5308977961540222, "lr": 0.0001, "grad_norm": 0.15364842307200952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89930116871, "tokens_per_second": 151912.3556869722, "data_wait_s": 1.3777998901787214e-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 sha256:5f1592850e24a456e9f8a0c7f8748f7ac936ac150b7e25ed5a5deed5c7dceedf
-size 211413
+oid sha256:34895ad7db98aeb88f749b2bc6fab35984ba8be50ac1a015dee894bec7e637b0
+size 213531