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": 74820.74824275514,
-      "required_pages_per_second_upper_estimate": 0.006485272826229289
+      "required_token_positions_per_second": 75901.45839172717,
+      "required_pages_per_second_upper_estimate": 0.0065789460429073535
     },
     {
       "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": 9352.593530344393,
-      "required_pages_per_second_upper_estimate": 0.0037308893929888275
+      "required_token_positions_per_second": 9487.682298965896,
+      "required_pages_per_second_upper_estimate": 0.003784778322549025
     },
     {
       "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": 18705.187060688786,
-      "required_pages_per_second_upper_estimate": 0.00432711582644166
+      "required_token_positions_per_second": 18975.36459793179,
+      "required_pages_per_second_upper_estimate": 0.004389616644720576
     },
     {
       "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": 28057.78059103318,
-      "required_pages_per_second_upper_estimate": 0.0017085716301959777
+      "required_token_positions_per_second": 28463.04689689769,
+      "required_pages_per_second_upper_estimate": 0.001733250221955146
     },
     {
       "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": 28057.78059103318,
-      "required_pages_per_second_upper_estimate": 0.0021599656188054416
+      "required_token_positions_per_second": 28463.04689689769,
+      "required_pages_per_second_upper_estimate": 0.0021911641408798276
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28057.78059103318,
-      "required_pages_per_second_upper_estimate": 0.34838373158961944
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28463.04689689769,
+      "required_pages_per_second_upper_estimate": 0.3534157827693816
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.08523219106257329,
-    "producer_p95_seconds": 0.279002989998844,
-    "gpu_update_seconds": 0.9853951174183698,
-    "producer_batches_per_second": 11.732656259720567,
-    "gpu_batches_per_second": 1.0148213466085496,
-    "producer_headroom": 11.56130219267672,
+    "producer_batch_seconds": 0.2505503376874856,
+    "producer_p95_seconds": 0.2789090709993616,
+    "gpu_update_seconds": 0.9713647347787449,
+    "producer_batches_per_second": 3.9912139382039546,
+    "gpu_batches_per_second": 1.0294794161204315,
+    "producer_headroom": 3.876924468528714,
     "sustainable": true,
-    "buffer_seconds": 60.10910216252056,
+    "buffer_seconds": 60.22461355628218,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 516854.65651620977,
-    "upload_bytes_per_second": 5669494.292678906,
+    "generation_bytes_per_second": 623561.611654566,
+    "upload_bytes_per_second": 5739472.249607404,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.90497752775879,
+    "estimated_full_shards_per_hour": 66.90090304483287,
     "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": 1.1727791039993463,
-      "bytes": 7506500,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.0041788279995671,
       "bytes": 4770921,
@@ -562,6 +556,12 @@
       "bytes": 1108747,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.292045906000567,
+      "bytes": 13335425,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -164,3 +164,4 @@
 {"step": 371400, "loss": 0.5784312561154366, "lr": 0.0003, "grad_norm": 0.16008242949486418, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54362101797, "tokens_per_second": 149632.86258804146, "data_wait_s": 8.464001439278945e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 371500, "loss": 0.6329603362828493, "lr": 0.0003, "grad_norm": 0.17358845159250366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54376784405, "tokens_per_second": 149154.9938721424, "data_wait_s": 1.4727000234415755e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 371600, "loss": 0.5862086229026318, "lr": 0.0003, "grad_norm": 0.18389443548017723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54391519551, "tokens_per_second": 150760.67801553255, "data_wait_s": 1.5630001144018024e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 371700, "loss": 0.6007288508117199, "lr": 0.0003, "grad_norm": 0.1356210092778565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54406064263, "tokens_per_second": 147811.33434016505, "data_wait_s": 8.936000085668638e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:530ee9c8ef00efb1a7499a1d2a3e636030593d2a2d41b61269219bbd84677c7d
-size 354102
+oid sha256:6d4f6ec762601f3d04a7ce3401f49e0576f91fcf9b6166c76a1d7a674d04d41b
+size 356220