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": 75934.49226841194,
-      "required_pages_per_second_upper_estimate": 0.006581809335614795
+      "required_token_positions_per_second": 75650.90195466953,
+      "required_pages_per_second_upper_estimate": 0.0065572284459726815
     },
     {
       "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": 9491.811533551492,
-      "required_pages_per_second_upper_estimate": 0.003786425535962778
+      "required_token_positions_per_second": 9456.362744333692,
+      "required_pages_per_second_upper_estimate": 0.0037722844839371676
     },
     {
       "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": 18983.623067102984,
-      "required_pages_per_second_upper_estimate": 0.004391527096219238
+      "required_token_positions_per_second": 18912.725488667384,
+      "required_pages_per_second_upper_estimate": 0.004375126189202927
     },
     {
       "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": 28475.434600654477,
-      "required_pages_per_second_upper_estimate": 0.0017340045681206756
+      "required_token_positions_per_second": 28369.088233001075,
+      "required_pages_per_second_upper_estimate": 0.0017275286322868534
     },
     {
       "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": 28475.434600654477,
-      "required_pages_per_second_upper_estimate": 0.0021921177806063837
+      "required_token_positions_per_second": 28369.088233001075,
+      "required_pages_per_second_upper_estimate": 0.002183930942838838
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28475.434600654477,
-      "required_pages_per_second_upper_estimate": 0.35356959659106346
+      "required_token_positions_per_second": 28369.088233001075,
+      "required_pages_per_second_upper_estimate": 0.3522491306232052
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.14343511235145456,
-    "producer_p95_seconds": 0.3482927869990817,
-    "gpu_update_seconds": 0.9709421607691473,
-    "producer_batches_per_second": 6.9717936118021875,
-    "gpu_batches_per_second": 1.0299274667482088,
-    "producer_headroom": 6.769208353879754,
+    "producer_batch_seconds": 0.11520756625785111,
+    "producer_p95_seconds": 0.33188998399782577,
+    "gpu_update_seconds": 0.9745819031236171,
+    "producer_batches_per_second": 8.679985459998834,
+    "gpu_batches_per_second": 1.02608102694593,
+    "producer_headroom": 8.45935674869099,
     "sustainable": true,
-    "buffer_seconds": 60.19841396768713,
+    "buffer_seconds": 60.42407799366426,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340903.15386649,
-    "upload_bytes_per_second": 5080432.717016629,
+    "generation_bytes_per_second": 369931.13623933174,
+    "upload_bytes_per_second": 5021018.832631182,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.14988100048089,
+    "estimated_full_shards_per_hour": 79.37861028084721,
     "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": 1.1651681650000683,
-      "bytes": 8287753,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1439757010011817,
       "bytes": 6251248,
@@ -562,6 +556,12 @@
       "bytes": 5900566,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0173681279993616,
+      "bytes": 2612410,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -317,3 +317,4 @@
 {"step": 386700, "loss": 0.6168648153543472, "lr": 0.0003, "grad_norm": 0.17218091571528626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56602285564, "tokens_per_second": 151670.51335805992, "data_wait_s": 9.251998562831432e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 386800, "loss": 0.544911116361618, "lr": 0.0003, "grad_norm": 0.1376656132044407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56616993030, "tokens_per_second": 149859.28636146747, "data_wait_s": 1.2904998584417626e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 386900, "loss": 0.6079707313328981, "lr": 0.0003, "grad_norm": 0.15555526425994035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56631539131, "tokens_per_second": 148225.39397405653, "data_wait_s": 1.2568998499773443e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 387000, "loss": 0.5514918360859156, "lr": 0.0003, "grad_norm": 0.1510423615634618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56646036656, "tokens_per_second": 146850.150214591, "data_wait_s": 9.527997463010252e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:802a70402fd5ba1b9ad241bc5bd0865d6d9b191c20000ba05bfc4d5a37558972
-size 680505
+oid sha256:df3c7f2b1a175a8cc11ed03017670153ac505d9f7cf628e6990d480a17e499de
+size 682623