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": 74457.37055768262,
-      "required_pages_per_second_upper_estimate": 0.006453776169459526
+      "required_token_positions_per_second": 74617.73909384164,
+      "required_pages_per_second_upper_estimate": 0.00646767650772345
     },
     {
       "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": 9307.171319710327,
-      "required_pages_per_second_upper_estimate": 0.00371276979404433
+      "required_token_positions_per_second": 9327.217386730204,
+      "required_pages_per_second_upper_estimate": 0.00372076646989397
     },
     {
       "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": 18614.342639420654,
-      "required_pages_per_second_upper_estimate": 0.004306100568388487
+      "required_token_positions_per_second": 18654.43477346041,
+      "required_pages_per_second_upper_estimate": 0.004315375178001123
     },
     {
       "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": 27921.51395913098,
-      "required_pages_per_second_upper_estimate": 0.001700273707248898
+      "required_token_positions_per_second": 27981.65216019061,
+      "required_pages_per_second_upper_estimate": 0.0017039358081726733
     },
     {
       "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": 27921.51395913098,
-      "required_pages_per_second_upper_estimate": 0.0021494754362714305
+      "required_token_positions_per_second": 27981.65216019061,
+      "required_pages_per_second_upper_estimate": 0.002154105041462196
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 27921.51395913098,
-      "required_pages_per_second_upper_estimate": 0.34669175607647396
+      "required_token_positions_per_second": 27981.65216019061,
+      "required_pages_per_second_upper_estimate": 0.3474384712640229
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.062050375460938945,
-    "producer_p95_seconds": 0.13404003300001932,
-    "gpu_update_seconds": 0.9902041859359301,
-    "producer_batches_per_second": 16.115937938675415,
-    "gpu_batches_per_second": 1.009892721322735,
-    "producer_headroom": 15.958069207160062,
+    "producer_batch_seconds": 0.07032776220311732,
+    "producer_p95_seconds": 0.1351400169999124,
+    "gpu_update_seconds": 0.9880760378879523,
+    "producer_batches_per_second": 14.21913578185308,
+    "gpu_batches_per_second": 1.0120678588031904,
+    "producer_headroom": 14.049587345524202,
     "sustainable": true,
-    "buffer_seconds": 60.402455342091734,
+    "buffer_seconds": 60.27263831116509,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 111,
+    "upload_pages_per_commit": 126,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 750323.560758434,
+    "generation_bytes_per_second": 682611.4412557194,
     "upload_bytes_per_second": 3229720.8444150747,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.50098475010283,
+    "estimated_full_shards_per_hour": 73.23626245619624,
     "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": [

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -3,3 +3,4 @@
 {"step": 82200, "loss": 0.5415969882160425, "lr": 0.0003, "grad_norm": 0.13570390417133701, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12029317961, "tokens_per_second": 149121.2710085824, "data_wait_s": 1.574399993842235e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 82300, "loss": 0.6576139517128468, "lr": 0.0003, "grad_norm": 0.12632208377102014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12044048685, "tokens_per_second": 150014.86912897477, "data_wait_s": 9.185999942928902e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 82400, "loss": 0.6455158451572061, "lr": 0.0003, "grad_norm": 0.15663461873576925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12058784343, "tokens_per_second": 148979.85896397586, "data_wait_s": 1.4244000112739741e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 82500, "loss": 0.6499202214181423, "lr": 0.0003, "grad_norm": 0.13765951160687467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12073486506, "tokens_per_second": 149389.11709760514, "data_wait_s": 8.5680001120636e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e13fb08de7ddcd2f7028419f08bef97e18ac6260119a9c9c75bc68e9587aa3bf
-size 10755
+oid sha256:f77ecb5f7197cc1a136c6640d86cda5d99d0c44b16d53a7ef11193edf56119ed
+size 12873