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": 76588.47363417498,
-      "required_pages_per_second_upper_estimate": 0.005298568315754681
+      "required_token_positions_per_second": 76810.27104273594,
+      "required_pages_per_second_upper_estimate": 0.0053139127751198195
     },
     {
       "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": 9573.559204271873,
-      "required_pages_per_second_upper_estimate": 0.00400824932017894
+      "required_token_positions_per_second": 9601.283880341993,
+      "required_pages_per_second_upper_estimate": 0.004019857063092428
     },
     {
       "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": 19147.118408543745,
-      "required_pages_per_second_upper_estimate": 0.003987839821031252
+      "required_token_positions_per_second": 19202.567760683985,
+      "required_pages_per_second_upper_estimate": 0.003999388458784312
     },
     {
       "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": 28720.67761281562,
-      "required_pages_per_second_upper_estimate": 0.001721760768952807
+      "required_token_positions_per_second": 28803.851641025976,
+      "required_pages_per_second_upper_estimate": 0.0017267469249446301
     },
     {
       "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": 28720.67761281562,
-      "required_pages_per_second_upper_estimate": 0.0022329063554864465
+      "required_token_positions_per_second": 28803.851641025976,
+      "required_pages_per_second_upper_estimate": 0.002239372770335915
     },
     {
       "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": 28720.67761281562,
-      "required_pages_per_second_upper_estimate": 0.3683980145561963
+      "required_token_positions_per_second": 28803.851641025976,
+      "required_pages_per_second_upper_estimate": 0.36946488168476516
     }
   ],
   "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": 0.534171398023318,
-    "producer_p95_seconds": 0.3223895970004378,
-    "gpu_update_seconds": 0.9626513821408944,
-    "producer_batches_per_second": 1.872058301325125,
-    "gpu_batches_per_second": 1.038797656713528,
-    "producer_headroom": 1.8021395112189666,
+    "producer_batch_seconds": 0.09143591564054532,
+    "producer_p95_seconds": 0.3203361699997913,
+    "gpu_update_seconds": 0.9598716291338041,
+    "producer_batches_per_second": 10.936621490522606,
+    "gpu_batches_per_second": 1.0418059765996086,
+    "producer_headroom": 10.497752687327706,
     "sustainable": true,
-    "buffer_seconds": 60.647037074876344,
+    "buffer_seconds": 60.47191263542966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 256162.04810976825,
-    "upload_bytes_per_second": 4734623.519998674,
+    "generation_bytes_per_second": 260705.9780971228,
+    "upload_bytes_per_second": 4803909.7269717455,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.96641237706934,
+    "estimated_full_shards_per_hour": 55.941433975079185,
     "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": 1.2197048900015943,
-      "bytes": 2772365,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.9406323449984484,
       "bytes": 5788082,
@@ -562,6 +556,12 @@
       "bytes": 2669509,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.7882141709997086,
+      "bytes": 11758267,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -395,3 +395,4 @@
 {"step": 759500, "loss": 0.5208806395530701, "lr": 5e-05, "grad_norm": 0.2359393559232186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111168881487, "tokens_per_second": 153367.73092413525, "data_wait_s": 1.2337004591245204e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 759600, "loss": 0.5625448226928711, "lr": 5e-05, "grad_norm": 0.24627748148452916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111183611984, "tokens_per_second": 140438.81696518476, "data_wait_s": 2.0908002625219524e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 759700, "loss": 0.5434607863426208, "lr": 5e-05, "grad_norm": 0.25023290452785857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111198298748, "tokens_per_second": 151753.92137527114, "data_wait_s": 1.4609002391807735e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 759800, "loss": 0.5847382545471191, "lr": 5e-05, "grad_norm": 0.26640136338412057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111212871325, "tokens_per_second": 151099.8639425124, "data_wait_s": 1.2169999536126852e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:369cee0764109e196ec4365f28b991ad08ffca283d99548ed572804a36240a9f
-size 846492
+oid sha256:20f71cf093268b6021e04e0120d32478097422b2ebbe6ca8eb60c31a135b8793
+size 848610