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": 75838.61411897183,
-      "required_pages_per_second_upper_estimate": 0.0065734988606222745
+      "required_token_positions_per_second": 75050.63518871983,
+      "required_pages_per_second_upper_estimate": 0.006505198844062371
     },
     {
       "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": 9479.82676487148,
-      "required_pages_per_second_upper_estimate": 0.003781644632548061
+      "required_token_positions_per_second": 9381.329398589978,
+      "required_pages_per_second_upper_estimate": 0.0037423525604715086
     },
     {
       "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": 18959.65352974296,
-      "required_pages_per_second_upper_estimate": 0.004385982165600446
+      "required_token_positions_per_second": 18762.658797179956,
+      "required_pages_per_second_upper_estimate": 0.004340410901210876
     },
     {
       "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": 28439.480294614437,
-      "required_pages_per_second_upper_estimate": 0.0017318151395205024
+      "required_token_positions_per_second": 28143.988195769936,
+      "required_pages_per_second_upper_estimate": 0.0017138212210280994
     },
     {
       "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": 28439.480294614437,
-      "required_pages_per_second_upper_estimate": 0.0021893499185995325
+      "required_token_positions_per_second": 28143.988195769936,
+      "required_pages_per_second_upper_estimate": 0.0021666021188560002
     },
     {
       "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": 28439.480294614437,
-      "required_pages_per_second_upper_estimate": 0.35312316444136777
+      "required_token_positions_per_second": 28143.988195769936,
+      "required_pages_per_second_upper_estimate": 0.3494541415221567
     }
   ],
   "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": 0.1036968103751974,
-    "producer_p95_seconds": 0.4056906369951321,
-    "gpu_update_seconds": 0.9721696639173706,
-    "producer_batches_per_second": 9.643498159507361,
-    "gpu_batches_per_second": 1.028627036118867,
-    "producer_headroom": 9.375116364716053,
+    "producer_batch_seconds": 0.15729041501560914,
+    "producer_p95_seconds": 0.4121766579992254,
+    "gpu_update_seconds": 0.9823767622300068,
+    "producer_batches_per_second": 6.357666485276692,
+    "gpu_batches_per_second": 1.0179393878678362,
+    "producer_headroom": 6.245623817144344,
     "sustainable": true,
-    "buffer_seconds": 60.27451916287698,
+    "buffer_seconds": 60.90735925826042,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 453490.8069030104,
-    "upload_bytes_per_second": 4586526.320916659,
+    "generation_bytes_per_second": 453461.29937621584,
+    "upload_bytes_per_second": 4616430.074456062,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.30857043569311,
+    "estimated_full_shards_per_hour": 97.30223880734307,
     "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": 0.9938246039964724,
-      "bytes": 235889,
-      "pages": 8,
-      "shards": 1
-    },
     {
       "seconds": 1.3448229619971244,
       "bytes": 7303500,
@@ -562,6 +556,12 @@
       "bytes": 6310458,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.6487007620016811,
+      "bytes": 5982194,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -427,3 +427,4 @@
 {"step": 197700, "loss": 0.6029997207224369, "lr": 0.0003, "grad_norm": 0.15610074409238134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28935300702, "tokens_per_second": 147006.29566422856, "data_wait_s": 8.101000275928527e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 197800, "loss": 0.5235050581395626, "lr": 0.0003, "grad_norm": 0.1499262392250177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28949997669, "tokens_per_second": 150175.95161636395, "data_wait_s": 9.081995813176036e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 197900, "loss": 0.5997746456414461, "lr": 0.0003, "grad_norm": 0.14653409194210884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28964748234, "tokens_per_second": 150073.80931078029, "data_wait_s": 1.5853001968935132e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 198000, "loss": 0.6337297223508358, "lr": 0.0003, "grad_norm": 0.17612064059506966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28979500934, "tokens_per_second": 149973.2657382262, "data_wait_s": 8.37100378703326e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2922e7ce0e560a1192ca34df3c62fafb53b390b68a5f23985def475d4abb7859
-size 915051
+oid sha256:1725c7038284eba96d7bbe3838c2bc4831294078532210220d9bfec8cb966124
+size 917169