Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75685.94793200468,
-      "required_pages_per_second_upper_estimate": 0.006560266142464875
+      "required_token_positions_per_second": 75157.17042322725,
+      "required_pages_per_second_upper_estimate": 0.00651443304817839
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9460.743491500585,
-      "required_pages_per_second_upper_estimate": 0.0037740320294800483
+      "required_token_positions_per_second": 9394.646302903406,
+      "required_pages_per_second_upper_estimate": 0.0037476648727075975
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18921.48698300117,
-      "required_pages_per_second_upper_estimate": 0.004377153006720014
+      "required_token_positions_per_second": 18789.29260580681,
+      "required_pages_per_second_upper_estimate": 0.004346572164097676
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28382.230474501757,
-      "required_pages_per_second_upper_estimate": 0.0017283289258422326
+      "required_token_positions_per_second": 28183.938908710217,
+      "required_pages_per_second_upper_estimate": 0.0017162540098409707
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28382.230474501757,
-      "required_pages_per_second_upper_estimate": 0.0021849426689696133
+      "required_token_positions_per_second": 28183.938908710217,
+      "required_pages_per_second_upper_estimate": 0.002169677635328795
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28382.230474501757,
-      "required_pages_per_second_upper_estimate": 0.3524123132783908
+      "required_token_positions_per_second": 28183.938908710217,
+      "required_pages_per_second_upper_estimate": 0.3499501956704399
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11774205185923847,
-    "producer_p95_seconds": 0.3507595939991006,
-    "gpu_update_seconds": 0.9741306281350446,
-    "producer_batches_per_second": 8.493142290364599,
-    "gpu_batches_per_second": 1.0265563684353933,
-    "producer_headroom": 8.273430034153177,
+    "producer_batch_seconds": 0.3413639609687493,
+    "producer_p95_seconds": 0.3689845970002352,
+    "gpu_update_seconds": 0.9809842438827957,
+    "producer_batches_per_second": 2.929424644482452,
+    "gpu_batches_per_second": 1.0193843644643452,
+    "producer_headroom": 2.8737194198792455,
     "sustainable": true,
-    "buffer_seconds": 60.39609894437277,
+    "buffer_seconds": 60.82102312073334,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356939.926097907,
+    "generation_bytes_per_second": 341581.6420395341,
     "upload_bytes_per_second": 6088898.701746191,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.59099900439175,
+    "estimated_full_shards_per_hour": 73.29546876801984,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -297,3 +297,4 @@
 {"step": 29700, "loss": 0.7051474638283253, "lr": 0.0003, "grad_norm": 0.14295477642627422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4346017764, "tokens_per_second": 148882.31375901506, "data_wait_s": 1.1593001545406878e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 29800, "loss": 0.6372893694788218, "lr": 0.0003, "grad_norm": 0.13709524730172185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4360541926, "tokens_per_second": 148383.01489362915, "data_wait_s": 1.182200139737688e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 29900, "loss": 0.7031157780438662, "lr": 0.0003, "grad_norm": 0.13538670995417385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4375163291, "tokens_per_second": 148461.46273434168, "data_wait_s": 1.149200033978559e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 30000, "loss": 0.6875663865357637, "lr": 0.0003, "grad_norm": 0.14237463414510249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4389888866, "tokens_per_second": 149914.74994791884, "data_wait_s": 9.903000318445265e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:67a1ef0095f281e63d9f3dad60659fbd29190fbf19c58739e3c270a81d9f3795
-size 631588
+oid sha256:b69d0e295289ad9fc337ca4a17d033b0234697d1ad3abd6a81c44b017205823e
+size 633706