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": 77251.07192117731,
-      "required_pages_per_second_upper_estimate": 0.0053444084026893456
+      "required_token_positions_per_second": 77151.93193928288,
+      "required_pages_per_second_upper_estimate": 0.005337549668705438
     },
     {
       "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": 9656.383990147164,
-      "required_pages_per_second_upper_estimate": 0.004042926328446719
+      "required_token_positions_per_second": 9643.99149241036,
+      "required_pages_per_second_upper_estimate": 0.004037737848429099
     },
     {
       "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": 19312.767980294328,
-      "required_pages_per_second_upper_estimate": 0.0040223402583539234
+      "required_token_positions_per_second": 19287.98298482072,
+      "required_pages_per_second_upper_estimate": 0.004017178197421056
     },
     {
       "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": 28969.15197044149,
-      "required_pages_per_second_upper_estimate": 0.0017366564272941016
+      "required_token_positions_per_second": 28931.974477231077,
+      "required_pages_per_second_upper_estimate": 0.0017344276933428826
     },
     {
       "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": 28969.15197044149,
-      "required_pages_per_second_upper_estimate": 0.0022522241438686624
+      "required_token_positions_per_second": 28931.974477231077,
+      "required_pages_per_second_upper_estimate": 0.0022493337572980633
     },
     {
       "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": 28969.15197044149,
-      "required_pages_per_second_upper_estimate": 0.37158517682484177
+      "required_token_positions_per_second": 28931.974477231077,
+      "required_pages_per_second_upper_estimate": 0.37110830386002075
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09579358207787436,
-    "producer_p95_seconds": 0.34209923199523473,
-    "gpu_update_seconds": 0.9543945238096884,
-    "producer_batches_per_second": 10.439112707854068,
-    "gpu_batches_per_second": 1.047784721153121,
-    "producer_headroom": 9.96303200180805,
+    "producer_batch_seconds": 0.11330320894563783,
+    "producer_p95_seconds": 0.32973902199591976,
+    "gpu_update_seconds": 0.955620917672192,
+    "producer_batches_per_second": 8.825875359627226,
+    "gpu_batches_per_second": 1.0464400490896657,
+    "producer_headroom": 8.434191110427356,
     "sustainable": true,
-    "buffer_seconds": 60.126855000010366,
+    "buffer_seconds": 60.2041178133481,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 459105.92501584993,
+    "generation_bytes_per_second": 472577.75103201496,
     "upload_bytes_per_second": 5297202.370957243,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.5134440694487,
+    "estimated_full_shards_per_hour": 101.40418432445847,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -378,3 +378,4 @@
 {"step": 434900, "loss": 0.5564521551132202, "lr": 0.0003, "grad_norm": 0.16359301725097797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63656864217, "tokens_per_second": 152202.06379904406, "data_wait_s": 1.2739001249428838e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 435000, "loss": 0.5770679116249084, "lr": 0.0003, "grad_norm": 0.15449644206745677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63671469675, "tokens_per_second": 152025.25322236057, "data_wait_s": 1.66549943969585e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 435100, "loss": 0.5555424690246582, "lr": 0.0003, "grad_norm": 0.16785926415375643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63686074299, "tokens_per_second": 142314.83712929187, "data_wait_s": 1.0607000149320811e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 435200, "loss": 0.6025435924530029, "lr": 0.0003, "grad_norm": 0.17389561632674724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63700561761, "tokens_per_second": 149297.00401588806, "data_wait_s": 1.2103002518415451e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c36a28407357bec663c205d90197750f2183aef6b2803972d386dc3b1609dbb
-size 811269
+oid sha256:f6b46bf84dbcf28ce510239d8151750eb6f764c800a54332e6b951c4b44ab313
+size 813387