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": 75860.27622960851,
-      "required_pages_per_second_upper_estimate": 0.006575376477470144
+      "required_token_positions_per_second": 73191.64784263867,
+      "required_pages_per_second_upper_estimate": 0.00634406653246441
     },
     {
       "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": 9482.534528701064,
-      "required_pages_per_second_upper_estimate": 0.0037827248000243596
+      "required_token_positions_per_second": 9148.955980329834,
+      "required_pages_per_second_upper_estimate": 0.0036496553296353255
     },
     {
       "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": 18965.06905740213,
-      "required_pages_per_second_upper_estimate": 0.0043872349526143635
+      "required_token_positions_per_second": 18297.91196065967,
+      "required_pages_per_second_upper_estimate": 0.004232899899846874
     },
     {
       "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": 28447.603586103192,
-      "required_pages_per_second_upper_estimate": 0.001732309805352551
+      "required_token_positions_per_second": 27446.867940989505,
+      "required_pages_per_second_upper_estimate": 0.0016713702550192808
     },
     {
       "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": 28447.603586103192,
-      "required_pages_per_second_upper_estimate": 0.002189975272065047
+      "required_token_positions_per_second": 27446.867940989505,
+      "required_pages_per_second_upper_estimate": 0.002112935871890628
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28447.603586103192,
-      "required_pages_per_second_upper_estimate": 0.3532240285347504
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27446.867940989505,
+      "required_pages_per_second_upper_estimate": 0.3407982410691919
     }
   ],
   "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": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11015087132034296,
-    "producer_p95_seconds": 0.362408161003259,
-    "gpu_update_seconds": 0.9718920581945326,
-    "producer_batches_per_second": 9.078457464869071,
-    "gpu_batches_per_second": 1.0289208472982927,
-    "producer_headroom": 8.82328071076312,
+    "producer_batch_seconds": 0.6340655790469896,
+    "producer_p95_seconds": 0.36197990799701074,
+    "gpu_update_seconds": 1.0073280513989038,
+    "producer_batches_per_second": 1.5771239333051568,
+    "gpu_batches_per_second": 0.9927252582823172,
+    "producer_headroom": 1.5886811785508583,
     "sustainable": true,
-    "buffer_seconds": 60.25730760806102,
+    "buffer_seconds": 60.43968308393423,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392314.48932748137,
-    "upload_bytes_per_second": 6153439.552201089,
+    "generation_bytes_per_second": 357043.9077039416,
+    "upload_bytes_per_second": 6115152.214670481,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.1815567957719,
+    "estimated_full_shards_per_hour": 76.61331103647886,
     "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": 1.1994012059985835,
-      "bytes": 10151059,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9821173639975314,
       "bytes": 2386369,
@@ -562,6 +556,12 @@
       "bytes": 1090867,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 0.8992618750016845,
+      "bytes": 5707571,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -282,3 +282,4 @@
 {"step": 28200, "loss": 0.6506574675440788, "lr": 0.0003, "grad_norm": 0.1307841212003055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4126800639, "tokens_per_second": 149813.64561755268, "data_wait_s": 1.1559000995475799e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 28300, "loss": 0.680478785187006, "lr": 0.0003, "grad_norm": 0.1456216874475559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4141315645, "tokens_per_second": 148029.4598356252, "data_wait_s": 1.1807002010755241e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 28400, "loss": 0.6338918879628181, "lr": 0.0003, "grad_norm": 0.1361738246486249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4155883368, "tokens_per_second": 148038.26469371165, "data_wait_s": 1.105399860534817e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 28500, "loss": 0.6514467559754848, "lr": 0.0003, "grad_norm": 0.13258373749732572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4170333192, "tokens_per_second": 128027.29278987453, "data_wait_s": 1.0427000233903527e-05, "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:bd3efac99644c01ae3af1dac450590d949ee28a9424056eac9abd90c99dc3088
-size 599818
+oid sha256:97418d1d0c9556d76c9c2c10aee96ac5649621ea39ea523b635aa824732b41aa
+size 601936