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": 77529.47961607407,
-      "required_pages_per_second_upper_estimate": 0.005363669293017156
+      "required_token_positions_per_second": 76833.05632696988,
+      "required_pages_per_second_upper_estimate": 0.0053154891139522665
     },
     {
       "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": 9691.184952009258,
-      "required_pages_per_second_upper_estimate": 0.0040574967644516555
+      "required_token_positions_per_second": 9604.132040871234,
+      "required_pages_per_second_upper_estimate": 0.004021049528429666
     },
     {
       "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": 19382.369904018517,
-      "required_pages_per_second_upper_estimate": 0.00403683650353951
+      "required_token_positions_per_second": 19208.26408174247,
+      "required_pages_per_second_upper_estimate": 0.004000574852238706
     },
     {
       "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": 29073.55485602777,
-      "required_pages_per_second_upper_estimate": 0.0017429152208710231
+      "required_token_positions_per_second": 28812.396122613707,
+      "required_pages_per_second_upper_estimate": 0.0017272591535691466
     },
     {
       "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": 29073.55485602777,
-      "required_pages_per_second_upper_estimate": 0.002260341009003234
+      "required_token_positions_per_second": 28812.396122613707,
+      "required_pages_per_second_upper_estimate": 0.0022400370661961586
     },
     {
       "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": 29073.55485602777,
-      "required_pages_per_second_upper_estimate": 0.37292434494205784
+      "required_token_positions_per_second": 28812.396122613707,
+      "required_pages_per_second_upper_estimate": 0.369574481120223
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08063231009381866,
-    "producer_p95_seconds": 0.19083142499948735,
-    "gpu_update_seconds": 0.950967301278185,
-    "producer_batches_per_second": 12.401976314909783,
-    "gpu_batches_per_second": 1.0515608671885044,
-    "producer_headroom": 11.793873946705725,
+    "producer_batch_seconds": 0.5186068377576447,
+    "producer_p95_seconds": 0.18743791899942153,
+    "gpu_update_seconds": 0.9595869736880434,
+    "producer_batches_per_second": 1.9282429910176384,
+    "gpu_batches_per_second": 1.0421150217959239,
+    "producer_headroom": 1.8503168562857966,
     "sustainable": true,
-    "buffer_seconds": 60.86190728180384,
+    "buffer_seconds": 60.453979342346734,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269707.430909094,
-    "upload_bytes_per_second": 4970249.090536304,
+    "generation_bytes_per_second": 231192.16223787254,
+    "upload_bytes_per_second": 5096493.83256463,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.872936205431124,
+    "estimated_full_shards_per_hour": 49.60845613815434,
     "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": 3.807216140000037,
-      "bytes": 7676406,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 2.506013934999828,
       "bytes": 5150101,
@@ -562,6 +556,12 @@
       "bytes": 6106895,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0866103590014973,
+      "bytes": 4773621,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -105,3 +105,4 @@
 {"step": 450700, "loss": 0.6282864212989807, "lr": 0.0003, "grad_norm": 0.16398398248758808, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65968400941, "tokens_per_second": 153336.29620067237, "data_wait_s": 1.0555999324424192e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 450800, "loss": 0.6025879383087158, "lr": 0.0003, "grad_norm": 0.15839335428853707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65983024556, "tokens_per_second": 152071.65040896164, "data_wait_s": 1.1902000551344827e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 450900, "loss": 0.6041690111160278, "lr": 0.0003, "grad_norm": 0.18410040394715915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65997771817, "tokens_per_second": 153932.7635396833, "data_wait_s": 1.0592999387881719e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 451000, "loss": 0.5745674967765808, "lr": 0.0003, "grad_norm": 0.20043337512395465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66012422465, "tokens_per_second": 127994.00135531738, "data_wait_s": 7.939001079648733e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0da3fb5800090c7dab8674c8df66aaf725f675f12966734d8c25fa0c267691c6
-size 228357
+oid sha256:b716d16d272e6003f857431356e01348d86246997836418040aeb4b0dc499a31
+size 230475