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": 75962.85593207141,
-      "required_pages_per_second_upper_estimate": 0.006584267826093749
+      "required_token_positions_per_second": 76048.98090599371,
+      "required_pages_per_second_upper_estimate": 0.006591732920551583
     },
     {
       "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": 9495.356991508927,
-      "required_pages_per_second_upper_estimate": 0.003787839872151319
+      "required_token_positions_per_second": 9506.122613249214,
+      "required_pages_per_second_upper_estimate": 0.0037921344396239084
     },
     {
       "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": 18990.713983017853,
-      "required_pages_per_second_upper_estimate": 0.0043931674548202855
+      "required_token_positions_per_second": 19012.245226498428,
+      "required_pages_per_second_upper_estimate": 0.004398148329062573
     },
     {
       "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": 28486.07097452678,
-      "required_pages_per_second_upper_estimate": 0.001734652267484758
+      "required_token_positions_per_second": 28518.36783974764,
+      "required_pages_per_second_upper_estimate": 0.0017366189771281524
     },
     {
       "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": 28486.07097452678,
-      "required_pages_per_second_upper_estimate": 0.0021929365981104413
+      "required_token_positions_per_second": 28518.36783974764,
+      "required_pages_per_second_upper_estimate": 0.002195422900198588
     },
     {
       "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": 28486.07097452678,
-      "required_pages_per_second_upper_estimate": 0.35370166475690407
+      "required_token_positions_per_second": 28518.36783974764,
+      "required_pages_per_second_upper_estimate": 0.3541026837322925
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.15311896103156641,
-    "producer_p95_seconds": 0.3609047920035664,
-    "gpu_update_seconds": 0.970579622044886,
-    "producer_batches_per_second": 6.530869810394311,
-    "gpu_batches_per_second": 1.0303121735578262,
-    "producer_headroom": 6.3387291521968665,
+    "producer_batch_seconds": 0.09654215730461146,
+    "producer_p95_seconds": 0.35586673200305086,
+    "gpu_update_seconds": 0.9694804469653218,
+    "producer_batches_per_second": 10.358169196952819,
+    "gpu_batches_per_second": 1.0314803182779095,
+    "producer_headroom": 10.042042502804247,
     "sustainable": true,
-    "buffer_seconds": 60.17593656678293,
+    "buffer_seconds": 60.10778771184995,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301015.54523045226,
+    "generation_bytes_per_second": 343007.33233068476,
     "upload_bytes_per_second": 5135396.880468336,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.5909287231939,
+    "estimated_full_shards_per_hour": 73.60138871612936,
     "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": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -392,3 +392,4 @@
 {"step": 276300, "loss": 0.572891453281045, "lr": 0.0003, "grad_norm": 0.20148212619272768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40442115845, "tokens_per_second": 149892.38877773884, "data_wait_s": 1.2851000064983964e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276400, "loss": 0.5967114940285683, "lr": 0.0003, "grad_norm": 0.15749853476536574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40456654049, "tokens_per_second": 148723.66149260214, "data_wait_s": 1.3636999938171357e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276500, "loss": 0.6283031068742275, "lr": 0.0003, "grad_norm": 0.15542361558506881, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40471107394, "tokens_per_second": 147453.32761461716, "data_wait_s": 1.09350003185682e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 276600, "loss": 0.580127177760005, "lr": 0.0003, "grad_norm": 0.14188547742678123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40485662186, "tokens_per_second": 142526.23012206602, "data_wait_s": 1.4037002983968705e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:31f967c3b49ae46e8a4838609bc9aad7ab8d34e0943f6b616aee857901699d6d
-size 840921
+oid sha256:b130a50d86f7cdde9ceefc2d0b0720cd4c59b7ef8b94617a67ff3daca03adbb9
+size 843039