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": 77386.23186445268,
-      "required_pages_per_second_upper_estimate": 0.005353759081179412
+      "required_token_positions_per_second": 76227.29150345013,
+      "required_pages_per_second_upper_estimate": 0.005273580897893131
     },
     {
       "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": 9673.278983056585,
-      "required_pages_per_second_upper_estimate": 0.004049999909170323
+      "required_token_positions_per_second": 9528.411437931267,
+      "required_pages_per_second_upper_estimate": 0.003989346893204698
     },
     {
       "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": 19346.55796611317,
-      "required_pages_per_second_upper_estimate": 0.004029377821298139
+      "required_token_positions_per_second": 19056.822875862534,
+      "required_pages_per_second_upper_estimate": 0.003969033642827084
     },
     {
       "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": 29019.836949169756,
-      "required_pages_per_second_upper_estimate": 0.0017396949143773798
+      "required_token_positions_per_second": 28585.2343137938,
+      "required_pages_per_second_upper_estimate": 0.0017136411499863917
     },
     {
       "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": 29019.836949169756,
-      "required_pages_per_second_upper_estimate": 0.0022561646780251245
+      "required_token_positions_per_second": 28585.2343137938,
+      "required_pages_per_second_upper_estimate": 0.0022223762347396113
     },
     {
       "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": 29019.836949169756,
-      "required_pages_per_second_upper_estimate": 0.37223530931067783
+      "required_token_positions_per_second": 28585.2343137938,
+      "required_pages_per_second_upper_estimate": 0.36666069334402845
     }
   ],
   "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": 0.06867588826551696,
-    "producer_p95_seconds": 0.18363794299875735,
-    "gpu_update_seconds": 0.9527276134744443,
-    "producer_batches_per_second": 14.561151304425323,
-    "gpu_batches_per_second": 1.0496179452101329,
-    "producer_headroom": 13.87281093170543,
+    "producer_batch_seconds": 0.1105881311171828,
+    "producer_p95_seconds": 0.21766171800118173,
+    "gpu_update_seconds": 0.9672126419008733,
+    "producer_batches_per_second": 9.04256170981285,
+    "gpu_batches_per_second": 1.033898810539417,
+    "producer_headroom": 8.746080000899763,
     "sustainable": true,
-    "buffer_seconds": 60.021839648889994,
+    "buffer_seconds": 60.93439643975502,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289940.9299605396,
-    "upload_bytes_per_second": 5460662.005188741,
+    "generation_bytes_per_second": 285381.67946809373,
+    "upload_bytes_per_second": 5553560.1583004,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.21457409011976,
+    "estimated_full_shards_per_hour": 61.23626506836042,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 5989028,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 1.483354228999815,
+      "bytes": 13516314,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -92,3 +92,4 @@
 {"step": 769200, "loss": 0.49378257989883423, "lr": 3e-05, "grad_norm": 0.26130389132820514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112588212168, "tokens_per_second": 152889.41558694514, "data_wait_s": 1.3338998542167246e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 769300, "loss": 0.5398075580596924, "lr": 3e-05, "grad_norm": 0.2597628498803907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112602760482, "tokens_per_second": 151855.2632847482, "data_wait_s": 1.214099938806612e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 769400, "loss": 0.48373109102249146, "lr": 3e-05, "grad_norm": 0.23279685805445827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112617267281, "tokens_per_second": 148437.4568293413, "data_wait_s": 1.5268000424839556e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 769500, "loss": 0.523493766784668, "lr": 3e-05, "grad_norm": 0.236160730364222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112631844636, "tokens_per_second": 151244.69218965745, "data_wait_s": 1.0974999895552173e-05, "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 sha256:6bdd0c8731c99f06b5b3fc447401652b04fcf631bdb72ded64d6d35890b8c169
-size 200040
+oid sha256:09c15bededc845c7a9079e566dfbca725c0a6c9d8e2a48cd5128203122bc1fc9
+size 202158