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": 76400.702468659,
-      "required_pages_per_second_upper_estimate": 0.005285577870835155
+      "required_token_positions_per_second": 75748.5152379929,
+      "required_pages_per_second_upper_estimate": 0.00524045804493481
     },
     {
       "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": 9550.087808582375,
-      "required_pages_per_second_upper_estimate": 0.003998422336942226
+      "required_token_positions_per_second": 9468.564404749113,
+      "required_pages_per_second_upper_estimate": 0.003964290190159497
     },
     {
       "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": 19100.17561716475,
-      "required_pages_per_second_upper_estimate": 0.003978062875551665
+      "required_token_positions_per_second": 18937.128809498226,
+      "required_pages_per_second_upper_estimate": 0.003944104525348198
     },
     {
       "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": 28650.263425747125,
-      "required_pages_per_second_upper_estimate": 0.0017175395459542889
+      "required_token_positions_per_second": 28405.693214247338,
+      "required_pages_per_second_upper_estimate": 0.0017028779352119669
     },
     {
       "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": 28650.263425747125,
-      "required_pages_per_second_upper_estimate": 0.002227431962161148
+      "required_token_positions_per_second": 28405.693214247338,
+      "required_pages_per_second_upper_estimate": 0.002208417703967711
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28650.263425747125,
-      "required_pages_per_second_upper_estimate": 0.3674948169693453
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28405.693214247338,
+      "required_pages_per_second_upper_estimate": 0.36435773289525963
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0745995468515872,
-    "producer_p95_seconds": 0.19039384700045048,
-    "gpu_update_seconds": 0.9650173050469608,
-    "producer_batches_per_second": 13.40490716370516,
-    "gpu_batches_per_second": 1.036250847285414,
-    "producer_headroom": 12.935967385523451,
+    "producer_batch_seconds": 0.05537231212503002,
+    "producer_p95_seconds": 0.1873168469992379,
+    "gpu_update_seconds": 0.9733260086795803,
+    "producer_batches_per_second": 18.05956734734161,
+    "gpu_batches_per_second": 1.027404991834756,
+    "producer_headroom": 17.577846604668082,
     "sustainable": true,
-    "buffer_seconds": 60.79609021795853,
+    "buffer_seconds": 60.346212538133976,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 89,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 277550.1790140697,
-    "upload_bytes_per_second": 5458592.488142992,
+    "generation_bytes_per_second": 300493.16611940745,
+    "upload_bytes_per_second": 5386211.437909366,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.555807378926936,
+    "estimated_full_shards_per_hour": 64.47883832632702,
     "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": [
@@ -490,6 +490,12 @@
       "bytes": 6141170,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0750775129999965,
+      "bytes": 1235188,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -98,3 +98,4 @@
 {"step": 769800, "loss": 0.5670958161354065, "lr": 3e-05, "grad_norm": 0.2698093247427889, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112676097216, "tokens_per_second": 153159.75086184766, "data_wait_s": 8.787999831838533e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 769900, "loss": 0.5111449360847473, "lr": 3e-05, "grad_norm": 0.24221454782284982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112690918043, "tokens_per_second": 154322.27194159306, "data_wait_s": 8.94799995876383e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 770000, "loss": 0.5791223049163818, "lr": 3e-05, "grad_norm": 0.28168012182084734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112705486715, "tokens_per_second": 150574.57346352158, "data_wait_s": 1.4241999451769516e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 770100, "loss": 0.5264490842819214, "lr": 3e-05, "grad_norm": 0.24843234261534192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112720043214, "tokens_per_second": 102502.96052938142, "data_wait_s": 1.5718000213382766e-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 sha256:161267f883c7078c0609bce739d9de167490d3f0fdf6a3588b7c7a27eb2fe9ec
-size 213531
+oid sha256:b060482a511d4fac153520fcbb7eceb25d41f5ce95dc534fce13a758b134603e
+size 215649