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": 75666.81508956345,
-      "required_pages_per_second_upper_estimate": 0.006558607756173815
+      "required_token_positions_per_second": 74008.42404693435,
+      "required_pages_per_second_upper_estimate": 0.006414862623752395
     },
     {
       "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": 9458.35188619543,
-      "required_pages_per_second_upper_estimate": 0.0037730779823661365
+      "required_token_positions_per_second": 9251.053005866794,
+      "required_pages_per_second_upper_estimate": 0.0036903833596085823
     },
     {
       "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": 18916.70377239086,
-      "required_pages_per_second_upper_estimate": 0.004376046495126951
+      "required_token_positions_per_second": 18502.106011733587,
+      "required_pages_per_second_upper_estimate": 0.004280136599870266
     },
     {
       "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": 28375.05565858629,
-      "required_pages_per_second_upper_estimate": 0.0017278920171963307
+      "required_token_positions_per_second": 27753.159017600385,
+      "required_pages_per_second_upper_estimate": 0.0016900217745997987
     },
     {
       "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": 28375.05565858629,
-      "required_pages_per_second_upper_estimate": 0.002184390331779279
+      "required_token_positions_per_second": 27753.159017600385,
+      "required_pages_per_second_upper_estimate": 0.0021365150068360993
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28375.05565858629,
-      "required_pages_per_second_upper_estimate": 0.35232322607728483
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27753.159017600385,
+      "required_pages_per_second_upper_estimate": 0.3446013511504077
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.16286537732077022,
-    "producer_p95_seconds": 0.44807584999944083,
-    "gpu_update_seconds": 0.97437694335002,
-    "producer_batches_per_second": 6.140040421423995,
-    "gpu_batches_per_second": 1.026296862651414,
-    "producer_headroom": 5.982713817872681,
+    "producer_batch_seconds": 0.5027136905075054,
+    "producer_p95_seconds": 0.42299423300573835,
+    "gpu_update_seconds": 0.9962109171956355,
+    "producer_batches_per_second": 1.9892038328824273,
+    "gpu_batches_per_second": 1.003803494560199,
+    "producer_headroom": 1.9816665748448765,
     "sustainable": true,
-    "buffer_seconds": 60.41137048770124,
+    "buffer_seconds": 60.768865948933765,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312346.7898394769,
-    "upload_bytes_per_second": 5398958.192868696,
+    "generation_bytes_per_second": 294587.5446804432,
+    "upload_bytes_per_second": 5473676.227065284,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.0223500384162,
+    "estimated_full_shards_per_hour": 63.211629441356386,
     "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": 0.8759991519982577,
-      "bytes": 805335,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.0561498970055254,
       "bytes": 10650758,
@@ -562,6 +556,12 @@
       "bytes": 943208,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.068753742001718,
+      "bytes": 7593652,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -406,3 +406,4 @@
 {"step": 355600, "loss": 0.5659326519817114, "lr": 0.0003, "grad_norm": 0.1694171388252178, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52049949793, "tokens_per_second": 149057.5964689041, "data_wait_s": 1.107400021282956e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 355700, "loss": 0.580079585313797, "lr": 0.0003, "grad_norm": 0.14113312019679025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52064543552, "tokens_per_second": 148225.1154835935, "data_wait_s": 9.800001862458885e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 355800, "loss": 0.6534725185483694, "lr": 0.0003, "grad_norm": 0.17635575235794385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52079015267, "tokens_per_second": 139459.0084785148, "data_wait_s": 1.3715995009988546e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 355900, "loss": 0.5479207634925842, "lr": 0.0003, "grad_norm": 0.14617998126550985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52093511172, "tokens_per_second": 147155.4351942816, "data_wait_s": 1.212100323755294e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d32e0896d7c49f2343f83347b1ab1a787a5b5a923ac064a8622d9544c596cdf
-size 870573
+oid sha256:9ba568b3617cb14554472c018d754d48e177468531e0bb3a11c28360714b15c9
+size 872691