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": 75583.20381247273,
-      "required_pages_per_second_upper_estimate": 0.006551360542586435
+      "required_token_positions_per_second": 75764.35483535356,
+      "required_pages_per_second_upper_estimate": 0.006567062254126676
     },
     {
       "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": 9447.90047655909,
-      "required_pages_per_second_upper_estimate": 0.003768908758799701
+      "required_token_positions_per_second": 9470.544354419195,
+      "required_pages_per_second_upper_estimate": 0.0037779417402342406
     },
     {
       "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": 18895.80095311818,
-      "required_pages_per_second_upper_estimate": 0.004371210995765271
+      "required_token_positions_per_second": 18941.08870883839,
+      "required_pages_per_second_upper_estimate": 0.004381687521013863
     },
     {
       "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": 28343.701429677272,
-      "required_pages_per_second_upper_estimate": 0.001725982709158697
+      "required_token_positions_per_second": 28411.63306325759,
+      "required_pages_per_second_upper_estimate": 0.0017301193892340018
     },
     {
       "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": 28343.701429677272,
-      "required_pages_per_second_upper_estimate": 0.002181976596443802
+      "required_token_positions_per_second": 28411.63306325759,
+      "required_pages_per_second_upper_estimate": 0.002187206161643612
     },
     {
       "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": 28343.701429677272,
-      "required_pages_per_second_upper_estimate": 0.3519339114900887
+      "required_token_positions_per_second": 28411.63306325759,
+      "required_pages_per_second_upper_estimate": 0.35277739502660377
     }
   ],
   "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": 0.4223004829687653,
-    "producer_p95_seconds": 0.2836854769993806,
-    "gpu_update_seconds": 0.9754548137827603,
-    "producer_batches_per_second": 2.367982136724109,
-    "gpu_batches_per_second": 1.0251628121266374,
-    "producer_headroom": 2.3098595742191184,
+    "producer_batch_seconds": 0.08184079783583798,
+    "producer_p95_seconds": 0.2165116409996699,
+    "gpu_update_seconds": 0.9731225212729806,
+    "producer_batches_per_second": 12.218844713682659,
+    "gpu_batches_per_second": 1.0276198301236106,
+    "producer_headroom": 11.8904329748219,
     "sustainable": true,
-    "buffer_seconds": 60.47819845453114,
+    "buffer_seconds": 60.3335963189248,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 247405.96499375624,
-    "upload_bytes_per_second": 6592326.588418496,
+    "generation_bytes_per_second": 247577.41918271518,
+    "upload_bytes_per_second": 6377105.426155326,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.08756077155605,
+    "estimated_full_shards_per_hour": 53.12435084925739,
     "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": 0.8824849620004898,
-      "bytes": 7297737,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.4247361650004677,
       "bytes": 17857886,
@@ -562,6 +556,12 @@
       "bytes": 12071070,
       "pages": 27,
       "shards": 4
+    },
+    {
+      "seconds": 3.228665867998643,
+      "bytes": 7676896,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -141,3 +141,4 @@
 {"step": 14100, "loss": 0.6691518537700176, "lr": 0.0003, "grad_norm": 0.15031498952481898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2063123218, "tokens_per_second": 149280.03400993772, "data_wait_s": 1.324299955740571e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 14200, "loss": 0.7174923568964005, "lr": 0.0003, "grad_norm": 0.14407498044637804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2077600356, "tokens_per_second": 147670.8799376657, "data_wait_s": 1.1600999641814269e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 14300, "loss": 0.6647667810320854, "lr": 0.0003, "grad_norm": 0.14701070813532818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2092167327, "tokens_per_second": 148300.1718091062, "data_wait_s": 1.3512999430531636e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 14400, "loss": 0.6537965591996908, "lr": 0.0003, "grad_norm": 0.13906760496417342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2106797171, "tokens_per_second": 149334.5009564062, "data_wait_s": 1.3601998944068328e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c9a5a9d15e7afc394224328fb1348e990feb90a7242370c46610e7c3b008aff
-size 299541
+oid sha256:31b4f84d09c640068d26b378c9946b80449b513297beff0d44d0f7978494359c
+size 301624