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": 77365.36100063048,
-      "required_pages_per_second_upper_estimate": 0.005352315186393116
+      "required_token_positions_per_second": 76388.95615836875,
+      "required_pages_per_second_upper_estimate": 0.005284765233834081
     },
     {
       "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": 9670.67012507881,
-      "required_pages_per_second_upper_estimate": 0.004048907634814178
+      "required_token_positions_per_second": 9548.619519796093,
+      "required_pages_per_second_upper_estimate": 0.0039978075950887655
     },
     {
       "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": 19341.34025015762,
-      "required_pages_per_second_upper_estimate": 0.00402829110866502
+      "required_token_positions_per_second": 19097.239039592187,
+      "required_pages_per_second_upper_estimate": 0.003977451263886058
     },
     {
       "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": 29012.010375236427,
-      "required_pages_per_second_upper_estimate": 0.0017392257232205631
+      "required_token_positions_per_second": 28645.858559388278,
+      "required_pages_per_second_upper_estimate": 0.0017172754809419175
     },
     {
       "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": 29012.010375236427,
-      "required_pages_per_second_upper_estimate": 0.002255556196327269
+      "required_token_positions_per_second": 28645.858559388278,
+      "required_pages_per_second_upper_estimate": 0.0022270895031767067
     },
     {
       "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": 29012.010375236427,
-      "required_pages_per_second_upper_estimate": 0.37213491842378144
+      "required_token_positions_per_second": 28645.858559388278,
+      "required_pages_per_second_upper_estimate": 0.367438316073271
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12834272639031497,
-    "producer_p95_seconds": 0.30695836700033396,
-    "gpu_update_seconds": 0.9529846309306198,
-    "producer_batches_per_second": 7.7916375015971475,
-    "gpu_batches_per_second": 1.049334866002475,
-    "producer_headroom": 7.425310788804734,
+    "producer_batch_seconds": 0.44724059650707204,
+    "producer_p95_seconds": 0.35053780000453116,
+    "gpu_update_seconds": 0.9651656955116381,
+    "producer_batches_per_second": 2.2359329806148027,
+    "gpu_batches_per_second": 1.0360915277556524,
+    "producer_headroom": 2.158045810352496,
     "sustainable": true,
-    "buffer_seconds": 60.038031748629045,
+    "buffer_seconds": 60.8054388172332,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 267394.09232677286,
-    "upload_bytes_per_second": 3828072.587352206,
+    "generation_bytes_per_second": 262687.04585132346,
+    "upload_bytes_per_second": 3836040.3860590337,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.376547597431085,
+    "estimated_full_shards_per_hour": 56.36652499823359,
     "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": 1.1358039750011812,
-      "bytes": 5944518,
-      "pages": 25,
-      "shards": 2
-    },
     {
       "seconds": 1.3701994980001473,
       "bytes": 11970289,
@@ -562,6 +556,12 @@
       "bytes": 1024661,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.288585857997532,
+      "bytes": 7415692,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -362,3 +362,4 @@
 {"step": 878700, "loss": 0.5303773880004883, "lr": 3e-05, "grad_norm": 0.32359358407629607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128617254746, "tokens_per_second": 150282.72802974845, "data_wait_s": 1.752200478222221e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 878800, "loss": 0.5227696299552917, "lr": 3e-05, "grad_norm": 0.31681058407222384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128631799245, "tokens_per_second": 151466.34347678034, "data_wait_s": 1.32610002765432e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 878900, "loss": 0.539548397064209, "lr": 3e-05, "grad_norm": 0.32667433365152343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128646391144, "tokens_per_second": 151418.38823903332, "data_wait_s": 1.35770023916848e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 879000, "loss": 0.560137152671814, "lr": 3e-05, "grad_norm": 0.3201839729736884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128661075387, "tokens_per_second": 128963.10808562308, "data_wait_s": 1.383299968438223e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:babc9fa61ae74440aeef1d7006d76a92b41a00eeff1f05875d30a07b33985b5b
-size 776598
+oid sha256:f6dddfd51443c5c588874bbac663dd0d2f3ef0bc57e80be17ffe15ecdf4c5ab6
+size 778716