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": 75479.8170578969,
-      "required_pages_per_second_upper_estimate": 0.005221868881370093
+      "required_token_positions_per_second": 76106.67581501119,
+      "required_pages_per_second_upper_estimate": 0.005265236424699969
     },
     {
       "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": 9434.977132237113,
-      "required_pages_per_second_upper_estimate": 0.003950227900540729
+      "required_token_positions_per_second": 9513.334476876398,
+      "required_pages_per_second_upper_estimate": 0.003983034484453774
     },
     {
       "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": 18869.954264474225,
-      "required_pages_per_second_upper_estimate": 0.003930113839131579
+      "required_token_positions_per_second": 19026.668953752796,
+      "required_pages_per_second_upper_estimate": 0.003962753376064027
     },
     {
       "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": 28304.931396711334,
-      "required_pages_per_second_upper_estimate": 0.0016968374181050704
+      "required_token_positions_per_second": 28540.003430629196,
+      "required_pages_per_second_upper_estimate": 0.0017109296276042335
     },
     {
       "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": 28304.931396711334,
-      "required_pages_per_second_upper_estimate": 0.0022005839158586826
+      "required_token_positions_per_second": 28540.003430629196,
+      "required_pages_per_second_upper_estimate": 0.002218859732523193
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28304.931396711334,
-      "required_pages_per_second_upper_estimate": 0.36306526848951826
+      "required_token_positions_per_second": 28540.003430629196,
+      "required_pages_per_second_upper_estimate": 0.3660805201399315
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06358799813274629,
-    "producer_p95_seconds": 0.21117167600095854,
-    "gpu_update_seconds": 0.9767909207231761,
-    "producer_batches_per_second": 15.726238116702467,
-    "gpu_batches_per_second": 1.023760539522256,
-    "producer_headroom": 15.36124660952571,
+    "producer_batch_seconds": 0.11379010199219408,
+    "producer_p95_seconds": 0.22606399100004637,
+    "gpu_update_seconds": 0.9687455037348771,
+    "producer_batches_per_second": 8.788110586882146,
+    "gpu_batches_per_second": 1.0322628555638453,
+    "producer_headroom": 8.513442617366952,
     "sustainable": true,
-    "buffer_seconds": 60.56103708483692,
+    "buffer_seconds": 60.06222123156238,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 439743.3895515994,
+    "generation_bytes_per_second": 440154.9621571735,
     "upload_bytes_per_second": 5646398.291855147,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.3586946955775,
+    "estimated_full_shards_per_hour": 94.44700859581378,
     "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

@@ -141,3 +141,4 @@
 {"step": 774100, "loss": 0.5496140718460083, "lr": 3e-05, "grad_norm": 0.24983376626042067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113305966819, "tokens_per_second": 151470.92394922976, "data_wait_s": 1.2017000699415803e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 774200, "loss": 0.4660772383213043, "lr": 3e-05, "grad_norm": 0.28941746315473155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113320583652, "tokens_per_second": 151429.46553302713, "data_wait_s": 1.1546000678208657e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 774300, "loss": 0.5189425945281982, "lr": 3e-05, "grad_norm": 0.25102008123817554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113335282156, "tokens_per_second": 152687.81659912478, "data_wait_s": 2.1437999748741277e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 774400, "loss": 0.5151251554489136, "lr": 3e-05, "grad_norm": 0.2618655086145514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113349957380, "tokens_per_second": 152664.69233507168, "data_wait_s": 1.537999924039468e-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:ab8a9084be4e4fdc4ce51e4a4489b703a9963d8f8de2cc79e8b502961fd09099
-size 304605
+oid sha256:11477b699428d08be2c5b56f85a05ceb8eebe1926bcd8fe708f58bd8a0843996
+size 306723