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": 75143.27492024105,
-      "required_pages_per_second_upper_estimate": 0.006513228621197377
+      "required_token_positions_per_second": 75357.1571411875,
+      "required_pages_per_second_upper_estimate": 0.006531767390029483
     },
     {
       "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": 9392.909365030131,
-      "required_pages_per_second_upper_estimate": 0.003746971982220413
+      "required_token_positions_per_second": 9419.644642648438,
+      "required_pages_per_second_upper_estimate": 0.003757637084190377
     },
     {
       "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": 18785.818730060262,
-      "required_pages_per_second_upper_estimate": 0.004345768544081838
+      "required_token_positions_per_second": 18839.289285296876,
+      "required_pages_per_second_upper_estimate": 0.004358138015986197
     },
     {
       "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": 28178.728095090395,
-      "required_pages_per_second_upper_estimate": 0.0017159366986305498
+      "required_token_positions_per_second": 28258.933927945316,
+      "required_pages_per_second_upper_estimate": 0.001720820813043931
     },
     {
       "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": 28178.728095090395,
-      "required_pages_per_second_upper_estimate": 0.0021692764924718857
+      "required_token_positions_per_second": 28258.933927945316,
+      "required_pages_per_second_upper_estimate": 0.002175450959509013
     },
     {
       "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": 28178.728095090395,
-      "required_pages_per_second_upper_estimate": 0.34988549480475306
+      "required_token_positions_per_second": 28258.933927945316,
+      "required_pages_per_second_upper_estimate": 0.3508813828171563
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1239068050865626,
+    "producer_batch_seconds": 0.4099975467348713,
     "producer_p95_seconds": 0.40487155500159133,
-    "gpu_update_seconds": 0.9811656476012889,
-    "producer_batches_per_second": 8.070581751352474,
-    "gpu_batches_per_second": 1.0191958946430264,
-    "producer_headroom": 7.918577570584894,
+    "gpu_update_seconds": 0.9783808572006617,
+    "producer_batches_per_second": 2.4390389844128975,
+    "gpu_batches_per_second": 1.0220968579262628,
+    "producer_headroom": 2.3863090523157218,
     "sustainable": true,
-    "buffer_seconds": 60.832270151279914,
+    "buffer_seconds": 60.659613146441025,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321861.3297646052,
-    "upload_bytes_per_second": 6292706.395301407,
+    "generation_bytes_per_second": 297897.2946797784,
+    "upload_bytes_per_second": 6282342.280009222,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.06394881919496,
+    "estimated_full_shards_per_hour": 63.921824744176995,
     "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": 1.1539005899976473,
-      "bytes": 5543356,
-      "pages": 19,
-      "shards": 2
-    },
     {
       "seconds": 0.995166409000376,
       "bytes": 7263909,
@@ -562,6 +556,12 @@
       "bytes": 11672731,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 0.8157727979996707,
+      "bytes": 2711312,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -381,3 +381,4 @@
 {"step": 78100, "loss": 0.6047646552324295, "lr": 0.0003, "grad_norm": 0.13520182350403326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11429812036, "tokens_per_second": 148336.8116493828, "data_wait_s": 1.0003001079894602e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 78200, "loss": 0.6694426946341991, "lr": 0.0003, "grad_norm": 0.14425529049851427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11444504104, "tokens_per_second": 150539.53375494335, "data_wait_s": 1.3578996004071087e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 78300, "loss": 0.5956331938505173, "lr": 0.0003, "grad_norm": 0.12783067811872012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11459096685, "tokens_per_second": 148629.83206661104, "data_wait_s": 1.3771001249551773e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 78400, "loss": 0.5999609120190144, "lr": 0.0003, "grad_norm": 0.1440686811274591, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11473586965, "tokens_per_second": 147650.72437113756, "data_wait_s": 1.316699490416795e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bfa05db6fc58c67e7d0266cf653dedd80f7fc8fd77b2c4e8009b6d7a898ae839
-size 816840
+oid sha256:b89a9ce1e768471c10078a8d7e4f836d2a7a2d77b95570a48aea56d0cc384303
+size 818958