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": 75380.00507323856,
-      "required_pages_per_second_upper_estimate": 0.006533747790872114
+      "required_token_positions_per_second": 74517.10660343278,
+      "required_pages_per_second_upper_estimate": 0.006458953938505527
     },
     {
       "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": 9422.50063415482,
-      "required_pages_per_second_upper_estimate": 0.003758776381903151
+      "required_token_positions_per_second": 9314.638325429098,
+      "required_pages_per_second_upper_estimate": 0.0037157484942672322
     },
     {
       "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": 18845.00126830964,
-      "required_pages_per_second_upper_estimate": 0.004359459382728732
+      "required_token_positions_per_second": 18629.276650858195,
+      "required_pages_per_second_upper_estimate": 0.004309555289104937
     },
     {
       "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": 28267.50190246446,
-      "required_pages_per_second_upper_estimate": 0.0017213425577394588
+      "required_token_positions_per_second": 27943.914976287295,
+      "required_pages_per_second_upper_estimate": 0.0017016378116646636
     },
     {
       "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": 28267.50190246446,
-      "required_pages_per_second_upper_estimate": 0.002176110545905168
+      "required_token_positions_per_second": 27943.914976287295,
+      "required_pages_per_second_upper_estimate": 0.002151199928582108
     },
     {
       "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": 28267.50190246446,
-      "required_pages_per_second_upper_estimate": 0.350987768385518
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27943.914976287295,
+      "required_pages_per_second_upper_estimate": 0.3469699017381737
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4024898483123849,
-    "producer_p95_seconds": 0.21706575200005318,
-    "gpu_update_seconds": 0.9780843066853937,
-    "producer_batches_per_second": 2.484534713590761,
-    "gpu_batches_per_second": 1.0224067528379797,
-    "producer_headroom": 2.430084412778213,
+    "producer_batch_seconds": 0.10688421684382376,
+    "producer_p95_seconds": 0.22493111299991142,
+    "gpu_update_seconds": 0.9894103966270151,
+    "producer_batches_per_second": 9.355918296722631,
+    "gpu_batches_per_second": 1.0107029432974282,
+    "producer_headroom": 9.256842832770287,
     "sustainable": true,
-    "buffer_seconds": 60.64122701449441,
+    "buffer_seconds": 60.35403419424792,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 237907.45351110725,
-    "upload_bytes_per_second": 4956588.409946015,
+    "generation_bytes_per_second": 231556.0728330354,
+    "upload_bytes_per_second": 5001493.338387428,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 51.0494013214103,
+    "estimated_full_shards_per_hour": 49.686542880471194,
     "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": 0.966999364000003,
-      "bytes": 6885855,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.0705818949998047,
       "bytes": 10769730,
@@ -562,6 +556,12 @@
       "bytes": 2128705,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.1122738659996685,
+      "bytes": 11663489,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -81,3 +81,4 @@
 {"step": 8100, "loss": 0.693482581526041, "lr": 0.0003, "grad_norm": 0.1611827912346656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1184819071, "tokens_per_second": 149585.46369521818, "data_wait_s": 9.639001291361637e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 8200, "loss": 0.7218140847980976, "lr": 0.0003, "grad_norm": 0.15468966826019376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1199413328, "tokens_per_second": 148992.5304126804, "data_wait_s": 7.2950006142491475e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 8300, "loss": 0.7129519097507, "lr": 0.0003, "grad_norm": 0.15017483257089176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1214133324, "tokens_per_second": 149061.4445244452, "data_wait_s": 1.0202000339631923e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 8400, "loss": 0.7150008901953697, "lr": 0.0003, "grad_norm": 0.172544872834465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1228789213, "tokens_per_second": 148404.7426836842, "data_wait_s": 1.5133000488276593e-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:cf6d11a6c2769c753a222c7457efe3d32b70896ad6ce3393182c96db618326ec
-size 173788
+oid sha256:16a673da289bc2aca25d54b1181fc26c42ababd587a4660cba79be8c28aed019
+size 175871