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": 74875.71790824445,
-      "required_pages_per_second_upper_estimate": 0.006490037457514559
+      "required_token_positions_per_second": 74981.76091022632,
+      "required_pages_per_second_upper_estimate": 0.006499228996162816
     },
     {
       "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": 9359.464738530556,
-      "required_pages_per_second_upper_estimate": 0.0037336304206680056
+      "required_token_positions_per_second": 9372.72011377829,
+      "required_pages_per_second_upper_estimate": 0.003738918188039848
     },
     {
       "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": 18718.929477061112,
-      "required_pages_per_second_upper_estimate": 0.00433029489261113
+      "required_token_positions_per_second": 18745.44022755658,
+      "required_pages_per_second_upper_estimate": 0.004336427688164984
     },
     {
       "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": 28078.394215591667,
-      "required_pages_per_second_upper_estimate": 0.0017098268917802597
+      "required_token_positions_per_second": 28118.160341334868,
+      "required_pages_per_second_upper_estimate": 0.0017122484402012832
     },
     {
       "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": 28078.394215591667,
-      "required_pages_per_second_upper_estimate": 0.0021615525126861186
+      "required_token_positions_per_second": 28118.160341334868,
+      "required_pages_per_second_upper_estimate": 0.002164613819125512
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28078.394215591667,
-      "required_pages_per_second_upper_estimate": 0.3486396838172724
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28118.160341334868,
+      "required_pages_per_second_upper_estimate": 0.3491334460103414
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10844793582809586,
-    "producer_p95_seconds": 0.25542130999929213,
-    "gpu_update_seconds": 0.9846716941044772,
-    "producer_batches_per_second": 9.221014603589417,
-    "gpu_batches_per_second": 1.0155669204134719,
-    "producer_headroom": 9.079672071078516,
+    "producer_batch_seconds": 0.06794500396873104,
+    "producer_p95_seconds": 0.21698853999987477,
+    "gpu_update_seconds": 0.9832791215489403,
+    "producer_batches_per_second": 14.717785585239053,
+    "gpu_batches_per_second": 1.0170052206790678,
+    "producer_headroom": 14.471691281399512,
     "sustainable": true,
-    "buffer_seconds": 60.06497334037311,
+    "buffer_seconds": 60.9633055360343,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328059.3726736873,
-    "upload_bytes_per_second": 6024587.62703304,
+    "generation_bytes_per_second": 328187.6361326363,
+    "upload_bytes_per_second": 6000508.069666131,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.39390454442943,
+    "estimated_full_shards_per_hour": 70.42142689689938,
     "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": 1.0060759970001527,
-      "bytes": 2526019,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1933680550000645,
       "bytes": 16188168,
@@ -562,6 +556,12 @@
       "bytes": 11368590,
       "pages": 44,
       "shards": 4
+    },
+    {
+      "seconds": 1.024016710000069,
+      "bytes": 910901,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 128400, "loss": 0.6278967410326004, "lr": 0.0003, "grad_norm": 0.13230083132296136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18791023917, "tokens_per_second": 147358.81289116075, "data_wait_s": 1.3410000974545255e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 128500, "loss": 0.5948844086378813, "lr": 0.0003, "grad_norm": 0.12773248528638248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18805528580, "tokens_per_second": 148391.19595577905, "data_wait_s": 8.979000995168462e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 128600, "loss": 0.5597027037292719, "lr": 0.0003, "grad_norm": 0.13727610489651007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18820138030, "tokens_per_second": 148792.39745070718, "data_wait_s": 7.991000529727899e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 128700, "loss": 0.6112298388034105, "lr": 0.0003, "grad_norm": 0.14444824213307159, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18834807250, "tokens_per_second": 149769.14422496923, "data_wait_s": 1.2307000361033715e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6c5186fe29e5c0cdb9bf18853a5b682b5b763ed0cf1c26904c8013585dc44500
-size 289779
+oid sha256:3b71dde39edd99965723791c64ecb2383563793f833060baed534dbc7b144188
+size 291897