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": 77102.9347819107,
-      "required_pages_per_second_upper_estimate": 0.005334159931669363
+      "required_token_positions_per_second": 76158.91623019558,
+      "required_pages_per_second_upper_estimate": 0.005268850537836889
     },
     {
       "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": 9637.866847738838,
-      "required_pages_per_second_upper_estimate": 0.004035173587602257
+      "required_token_positions_per_second": 9519.864528774448,
+      "required_pages_per_second_upper_estimate": 0.003985768480820497
     },
     {
       "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": 19275.733695477677,
-      "required_pages_per_second_upper_estimate": 0.004014626993486383
+      "required_token_positions_per_second": 19039.729057548895,
+      "required_pages_per_second_upper_estimate": 0.003965473451266657
     },
     {
       "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": 28913.600543216515,
-      "required_pages_per_second_upper_estimate": 0.0017333262040540823
+      "required_token_positions_per_second": 28559.59358632334,
+      "required_pages_per_second_upper_estimate": 0.0017121040275256615
     },
     {
       "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": 28913.600543216515,
-      "required_pages_per_second_upper_estimate": 0.0022479052647468257
+      "required_token_positions_per_second": 28559.59358632334,
+      "required_pages_per_second_upper_estimate": 0.002220382780960424
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28913.600543216515,
-      "required_pages_per_second_upper_estimate": 0.3708726227628752
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28559.59358632334,
+      "required_pages_per_second_upper_estimate": 0.3663318016229055
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05493818685165763,
-    "producer_p95_seconds": 0.18927947399970435,
-    "gpu_update_seconds": 0.9562281929805023,
-    "producer_batches_per_second": 18.20227527166429,
-    "gpu_batches_per_second": 1.0457754826105512,
-    "producer_headroom": 17.405528791157227,
+    "producer_batch_seconds": 0.07375065738281705,
+    "producer_p95_seconds": 0.18886795200160122,
+    "gpu_update_seconds": 0.968081002848728,
+    "producer_batches_per_second": 13.559201171717108,
+    "gpu_batches_per_second": 1.0329714115423663,
+    "producer_headroom": 13.126405068143546,
     "sustainable": true,
-    "buffer_seconds": 60.242376157771645,
+    "buffer_seconds": 60.02102217662114,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292614.0560326726,
-    "upload_bytes_per_second": 5487441.673117563,
+    "generation_bytes_per_second": 276138.73139368935,
+    "upload_bytes_per_second": 5555397.99285478,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.78816471800931,
+    "estimated_full_shards_per_hour": 59.2529435764123,
     "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": [
@@ -472,6 +472,12 @@
       "bytes": 838138,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.1489447340009065,
+      "bytes": 10403004,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -95,3 +95,4 @@
 {"step": 769500, "loss": 0.523493766784668, "lr": 3e-05, "grad_norm": 0.236160730364222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112631844636, "tokens_per_second": 151244.69218965745, "data_wait_s": 1.0974999895552173e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 769600, "loss": 0.521182119846344, "lr": 3e-05, "grad_norm": 0.2844613332504851, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112646639597, "tokens_per_second": 154169.4638959554, "data_wait_s": 9.65899926086422e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 769700, "loss": 0.5391861200332642, "lr": 3e-05, "grad_norm": 0.24925319598414888, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112661368981, "tokens_per_second": 129550.51739029269, "data_wait_s": 1.6445999790448695e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 769800, "loss": 0.5670958161354065, "lr": 3e-05, "grad_norm": 0.2698093247427889, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112676097216, "tokens_per_second": 153159.75086184766, "data_wait_s": 8.787999831838533e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b36f3049758f5eab22c19e0684b043598eb23b2b71cebde4be64370eb5dce522
-size 206394
+oid sha256:02b6c011d01ff942cf0ba1ba06e7feecb6b2bc2a3fa734b87c6c2231fbc19160
+size 208512