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": 77156.6150675723,
-      "required_pages_per_second_upper_estimate": 0.005337873658386855
+      "required_token_positions_per_second": 77241.04553529725,
+      "required_pages_per_second_upper_estimate": 0.005343714754049743
     },
     {
       "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": 9644.576883446538,
-      "required_pages_per_second_upper_estimate": 0.004037982939431592
+      "required_token_positions_per_second": 9655.130691912156,
+      "required_pages_per_second_upper_estimate": 0.004042401598647564
     },
     {
       "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": 19289.153766893076,
-      "required_pages_per_second_upper_estimate": 0.004017422040451128
+      "required_token_positions_per_second": 19310.261383824312,
+      "required_pages_per_second_upper_estimate": 0.0040218182004126135
     },
     {
       "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": 28933.730650339614,
-      "required_pages_per_second_upper_estimate": 0.0017345329732392182
+      "required_token_positions_per_second": 28965.39207573647,
+      "required_pages_per_second_upper_estimate": 0.0017364310273475622
     },
     {
       "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": 28933.730650339614,
-      "required_pages_per_second_upper_estimate": 0.0022494702920326628
+      "required_token_positions_per_second": 28965.39207573647,
+      "required_pages_per_second_upper_estimate": 0.0022519318285933758
     },
     {
       "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": 28933.730650339614,
-      "required_pages_per_second_upper_estimate": 0.3711308301630253
+      "required_token_positions_per_second": 28965.39207573647,
+      "required_pages_per_second_upper_estimate": 0.3715369489326262
     }
   ],
   "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": 0.37896600327334795,
-    "producer_p95_seconds": 0.23710926200146787,
-    "gpu_update_seconds": 0.9555629149286865,
-    "producer_batches_per_second": 2.6387591271048674,
-    "gpu_batches_per_second": 1.0465035680823065,
-    "producer_headroom": 2.5215003632910036,
+    "producer_batch_seconds": 0.22742202159378166,
+    "producer_p95_seconds": 0.24827257400102098,
+    "gpu_update_seconds": 0.954518410374289,
+    "producer_batches_per_second": 4.397111559346646,
+    "gpu_batches_per_second": 1.0476487295911627,
+    "producer_headroom": 4.197123935865972,
     "sustainable": true,
-    "buffer_seconds": 60.20046364050725,
+    "buffer_seconds": 60.1346598535802,
     "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": 441218.51279274194,
-    "upload_bytes_per_second": 5219095.352802215,
+    "generation_bytes_per_second": 425533.1197608066,
+    "upload_bytes_per_second": 5295400.822475007,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.6752218040151,
+    "estimated_full_shards_per_hour": 91.30950159662389,
     "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": 0.7895571030003339,
-      "bytes": 1089032,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.843757688999176,
       "bytes": 12452068,
@@ -562,6 +556,12 @@
       "bytes": 6298065,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.972795835998113,
+      "bytes": 8254778,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -215,3 +215,4 @@
 {"step": 578600, "loss": 0.5609111189842224, "lr": 0.0003, "grad_norm": 0.17871318168536346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84689857947, "tokens_per_second": 151580.5485364575, "data_wait_s": 9.054001566255465e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 578700, "loss": 0.6683929562568665, "lr": 0.0003, "grad_norm": 0.18126979748246194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84704366090, "tokens_per_second": 150255.4647987151, "data_wait_s": 1.0960000508930534e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 578800, "loss": 0.709212064743042, "lr": 0.0003, "grad_norm": 0.16372910982491218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84719087562, "tokens_per_second": 154262.92936551804, "data_wait_s": 1.0092000593431294e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 578900, "loss": 0.581824004650116, "lr": 0.0003, "grad_norm": 0.16481607578175492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84733783930, "tokens_per_second": 152618.67735907383, "data_wait_s": 1.02440026239492e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0670c4220e2ed2c94e2a2d67778758a016cc20a35a92d0807e0a037b185de2ad
-size 462903
+oid sha256:fc3b3f2663846f52df08ff3c56f756fd69db0540acdf773fe8fb146c2a892c83
+size 465021