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": 76973.9701103644,
-      "required_pages_per_second_upper_estimate": 0.005325237856452524
+      "required_token_positions_per_second": 76636.89200181136,
+      "required_pages_per_second_upper_estimate": 0.005301918010773857
     },
     {
       "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": 9621.74626379555,
-      "required_pages_per_second_upper_estimate": 0.0040284242357412755
+      "required_token_positions_per_second": 9579.61150022642,
+      "required_pages_per_second_upper_estimate": 0.004010783290108798
     },
     {
       "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": 19243.4925275911,
-      "required_pages_per_second_upper_estimate": 0.004007912008472384
+      "required_token_positions_per_second": 19159.22300045284,
+      "required_pages_per_second_upper_estimate": 0.00399036088830636
     },
     {
       "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": 28865.23879138665,
-      "required_pages_per_second_upper_estimate": 0.0017304269908760014
+      "required_token_positions_per_second": 28738.834500679255,
+      "required_pages_per_second_upper_estimate": 0.0017228492466562692
     },
     {
       "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": 28865.23879138665,
-      "required_pages_per_second_upper_estimate": 0.0022441453512629193
+      "required_token_positions_per_second": 28738.834500679255,
+      "required_pages_per_second_upper_estimate": 0.0022343179736541347
     },
     {
       "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": 28865.23879138665,
-      "required_pages_per_second_upper_estimate": 0.3702522901372052
+      "required_token_positions_per_second": 28738.834500679255,
+      "required_pages_per_second_upper_estimate": 0.36863091161836375
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.1440605627812488,
-    "producer_p95_seconds": 0.3342231630012975,
-    "gpu_update_seconds": 0.9578302885285718,
-    "producer_batches_per_second": 6.941525013465808,
-    "gpu_batches_per_second": 1.0440262873042048,
-    "producer_headroom": 6.648802906476254,
+    "producer_batch_seconds": 0.283036707687927,
+    "producer_p95_seconds": 0.34494934599933913,
+    "gpu_update_seconds": 0.9620431893070168,
+    "producer_batches_per_second": 3.533110627836261,
+    "gpu_batches_per_second": 1.0394543728544292,
+    "producer_headroom": 3.399005016578113,
     "sustainable": true,
-    "buffer_seconds": 60.343308177300024,
+    "buffer_seconds": 60.60872092634206,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395977.7234001982,
-    "upload_bytes_per_second": 5405775.629479799,
+    "generation_bytes_per_second": 381972.02432326536,
+    "upload_bytes_per_second": 5374959.395048698,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.96760155205212,
+    "estimated_full_shards_per_hour": 81.96230456613036,
     "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": 1.2522269889996096,
-      "bytes": 10696457,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.1242637430004834,
       "bytes": 7345030,
@@ -562,6 +556,12 @@
       "bytes": 11581137,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.0408890239996254,
+      "bytes": 7081099,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -351,3 +351,4 @@
 {"step": 432200, "loss": 0.6296667456626892, "lr": 0.0003, "grad_norm": 0.18727836559296798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63262069641, "tokens_per_second": 150961.77623328965, "data_wait_s": 1.3265998859424144e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 432300, "loss": 0.5761712789535522, "lr": 0.0003, "grad_norm": 0.17810040164809454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63276586995, "tokens_per_second": 150486.99126934007, "data_wait_s": 1.3479999324772507e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 432400, "loss": 0.5923108458518982, "lr": 0.0003, "grad_norm": 0.1772879380050161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63291303295, "tokens_per_second": 152364.13310560162, "data_wait_s": 1.901299401652068e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 432500, "loss": 0.5722652673721313, "lr": 0.0003, "grad_norm": 0.1537789846340482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63305966357, "tokens_per_second": 152138.76700763498, "data_wait_s": 1.7398000636603683e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0d4bdf8a6671e135bbca0cf10a3fd540595a719506c08e1dbfd7a9defb83b186
-size 753300
+oid sha256:41478a6d6f817ec6aef22ddca4f4402b1b35981ce2d1c371a589321fd849d2c5
+size 755418