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": 75719.48362746985,
-      "required_pages_per_second_upper_estimate": 0.0065631729315523525
+      "required_token_positions_per_second": 75528.42842105473,
+      "required_pages_per_second_upper_estimate": 0.006546612750485284
     },
     {
       "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": 9464.93545343373,
-      "required_pages_per_second_upper_estimate": 0.0037757042657705965
+      "required_token_positions_per_second": 9441.053552631842,
+      "required_pages_per_second_upper_estimate": 0.0037661774184744863
     },
     {
       "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": 18929.87090686746,
-      "required_pages_per_second_upper_estimate": 0.00437909247995446
+      "required_token_positions_per_second": 18882.107105263683,
+      "required_pages_per_second_upper_estimate": 0.004368043165067624
     },
     {
       "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": 28394.80636030119,
-      "required_pages_per_second_upper_estimate": 0.0017290947313068448
+      "required_token_positions_per_second": 28323.160657895525,
+      "required_pages_per_second_upper_estimate": 0.0017247318839262893
     },
     {
       "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": 28394.80636030119,
-      "required_pages_per_second_upper_estimate": 0.0021859107954707337
+      "required_token_positions_per_second": 28323.160657895525,
+      "required_pages_per_second_upper_estimate": 0.0021803953109715475
     },
     {
       "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": 28394.80636030119,
-      "required_pages_per_second_upper_estimate": 0.35256846369123745
+      "required_token_positions_per_second": 28323.160657895525,
+      "required_pages_per_second_upper_estimate": 0.3516788638501003
     }
   ],
   "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": 0.13170573854694112,
-    "producer_p95_seconds": 0.34831243899679976,
-    "gpu_update_seconds": 0.9736991916470576,
-    "producer_batches_per_second": 7.5926835917145015,
-    "gpu_batches_per_second": 1.0270112254159864,
-    "producer_headroom": 7.392989875684289,
+    "producer_batch_seconds": 0.3162141930546625,
+    "producer_p95_seconds": 0.35135073899800773,
+    "gpu_update_seconds": 0.9761622417056299,
+    "producer_batches_per_second": 3.1624133956160994,
+    "gpu_batches_per_second": 1.024419873332448,
+    "producer_headroom": 3.0870285494645247,
     "sustainable": true,
-    "buffer_seconds": 60.369349882117575,
+    "buffer_seconds": 60.522058985749055,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396049.5212092127,
-    "upload_bytes_per_second": 3919643.0195153127,
+    "generation_bytes_per_second": 369978.05451473553,
+    "upload_bytes_per_second": 3925609.2230247515,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.98300769049918,
+    "estimated_full_shards_per_hour": 79.38867785054731,
     "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": 1.3117401380004594,
-      "bytes": 7488136,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9632922999990114,
       "bytes": 4570486,
@@ -562,6 +556,12 @@
       "bytes": 5985007,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1093607039983908,
+      "bytes": 7338185,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -235,3 +235,4 @@
 {"step": 138500, "loss": 0.5999759268015623, "lr": 0.0003, "grad_norm": 0.14957302639950082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20269059703, "tokens_per_second": 147830.30864382494, "data_wait_s": 1.2942000466864556e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 138600, "loss": 0.5904894862323999, "lr": 0.0003, "grad_norm": 0.1377562655743542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20283675767, "tokens_per_second": 149458.7950497035, "data_wait_s": 1.112900281441398e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 138700, "loss": 0.5369933377951384, "lr": 0.0003, "grad_norm": 0.13935597412628523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20298343512, "tokens_per_second": 148685.80107937855, "data_wait_s": 9.620001947041601e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 138800, "loss": 0.5648066718131304, "lr": 0.0003, "grad_norm": 0.15375385803934544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20313041480, "tokens_per_second": 150122.57332411828, "data_wait_s": 1.0962998203467578e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3902f7923c1cb2ccb628d8db83a60be0485e3872387a9f8faab17b00719441e
-size 505263
+oid sha256:a08cfeedf49802e9af9afb7842de21616499477e79f24d7e5e424f76b66f11a7
+size 507381