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": 76493.58146360646,
-      "required_pages_per_second_upper_estimate": 0.00529200345521458
+      "required_token_positions_per_second": 77289.59669854419,
+      "required_pages_per_second_upper_estimate": 0.005347073636177386
     },
     {
       "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": 9561.697682950808,
-      "required_pages_per_second_upper_estimate": 0.004003283148898542
+      "required_token_positions_per_second": 9661.199587318024,
+      "required_pages_per_second_upper_estimate": 0.004044942518421054
     },
     {
       "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": 19123.395365901615,
-      "required_pages_per_second_upper_estimate": 0.003982898936867601
+      "required_token_positions_per_second": 19322.399174636048,
+      "required_pages_per_second_upper_estimate": 0.004024346182143629
     },
     {
       "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": 28685.093048852425,
-      "required_pages_per_second_upper_estimate": 0.0017196275286776965
+      "required_token_positions_per_second": 28983.598761954076,
+      "required_pages_per_second_upper_estimate": 0.0017375224903863078
     },
     {
       "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": 28685.093048852425,
-      "required_pages_per_second_upper_estimate": 0.0022301398121582633
+      "required_token_positions_per_second": 28983.598761954076,
+      "required_pages_per_second_upper_estimate": 0.002253347318364046
     },
     {
       "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": 28685.093048852425,
-      "required_pages_per_second_upper_estimate": 0.3679415739774044
+      "required_token_positions_per_second": 28983.598761954076,
+      "required_pages_per_second_upper_estimate": 0.37177048475460905
     }
   ],
   "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": 0.7715019532265046,
-    "producer_p95_seconds": 0.2843123849997937,
-    "gpu_update_seconds": 0.9638455748745108,
-    "producer_batches_per_second": 1.2961729984193713,
-    "gpu_batches_per_second": 1.0375105992785165,
-    "producer_headroom": 1.2493106087983372,
+    "producer_batch_seconds": 0.11051528506251884,
+    "producer_p95_seconds": 0.29845391900016693,
+    "gpu_update_seconds": 0.9539188086019437,
+    "producer_batches_per_second": 9.048522106551115,
+    "gpu_batches_per_second": 1.0483072468878065,
+    "producer_headroom": 8.63155542748959,
     "sustainable": true,
-    "buffer_seconds": 60.72227121709418,
+    "buffer_seconds": 60.096884941922454,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317227.7540059514,
-    "upload_bytes_per_second": 4676002.18835287,
+    "generation_bytes_per_second": 305276.10750327137,
+    "upload_bytes_per_second": 4674474.790950291,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.06969132551104,
+    "estimated_full_shards_per_hour": 65.50514620612722,
     "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": 1.42488349499763,
-      "bytes": 9837963,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.2818151600004057,
       "bytes": 7876217,
@@ -562,6 +556,12 @@
       "bytes": 6710244,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.629695798998,
+      "bytes": 10655327,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -289,3 +289,4 @@
 {"step": 871400, "loss": 0.5220195651054382, "lr": 3e-05, "grad_norm": 0.3010914590787029, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127548657889, "tokens_per_second": 152149.5056921441, "data_wait_s": 1.077799970516935e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 871500, "loss": 0.49482545256614685, "lr": 3e-05, "grad_norm": 0.3150574178858843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127563145671, "tokens_per_second": 150970.0307476434, "data_wait_s": 1.9135997717967257e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 871600, "loss": 0.508460521697998, "lr": 3e-05, "grad_norm": 0.2952682373814299, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127577619184, "tokens_per_second": 117957.411074345, "data_wait_s": 8.72599775902927e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 871700, "loss": 0.5211670398712158, "lr": 3e-05, "grad_norm": 0.300449272457701, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127592222603, "tokens_per_second": 151054.1694849696, "data_wait_s": 1.438899926142767e-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:e703be13740ddc1171b0a4170e916730045c03a2338341b2bf163e2130c71bda
-size 621201
+oid sha256:b71df1463f9c2c15b17e4ce0e0c65e7463bec8776e2ca21417731a15dbd7900b
+size 623319