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": 74555.64485033134,
-      "required_pages_per_second_upper_estimate": 0.0064622943360724
+      "required_token_positions_per_second": 75029.94973643213,
+      "required_pages_per_second_upper_estimate": 0.006503405881484872
     },
     {
       "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": 9319.455606291418,
-      "required_pages_per_second_upper_estimate": 0.003717670179627979
+      "required_token_positions_per_second": 9378.743717054016,
+      "required_pages_per_second_upper_estimate": 0.0037413210934474296
     },
     {
       "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": 18638.911212582836,
-      "required_pages_per_second_upper_estimate": 0.004311784075397437
+      "required_token_positions_per_second": 18757.487434108032,
+      "required_pages_per_second_upper_estimate": 0.004339214597377068
     },
     {
       "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": 27958.366818874256,
-      "required_pages_per_second_upper_estimate": 0.0017025178530552514
+      "required_token_positions_per_second": 28136.23115116205,
+      "required_pages_per_second_upper_estimate": 0.0017133488577095468
     },
     {
       "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": 27958.366818874256,
-      "required_pages_per_second_upper_estimate": 0.002152312471429713
+      "required_token_positions_per_second": 28136.23115116205,
+      "required_pages_per_second_upper_estimate": 0.0021660049600892113
     },
     {
       "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": 27958.366818874256,
-      "required_pages_per_second_upper_estimate": 0.34714934525589797
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28136.23115116205,
+      "required_pages_per_second_upper_estimate": 0.349357824989285
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08956438560949209,
-    "producer_p95_seconds": 0.27786929499779944,
-    "gpu_update_seconds": 0.9888989646324859,
-    "producer_batches_per_second": 11.165152233166431,
-    "gpu_batches_per_second": 1.0112256517243292,
-    "producer_headroom": 11.041207483342372,
+    "producer_batch_seconds": 0.07483270482040894,
+    "producer_p95_seconds": 0.25393821399848093,
+    "gpu_update_seconds": 0.9826475995118527,
+    "producer_batches_per_second": 13.36314118806611,
+    "gpu_batches_per_second": 1.0176588234650625,
+    "producer_headroom": 13.13125861039113,
     "sustainable": true,
-    "buffer_seconds": 60.322836842581644,
+    "buffer_seconds": 60.92415116973486,
     "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": 302679.6693812885,
+    "generation_bytes_per_second": 302760.57776660786,
     "upload_bytes_per_second": 5474544.052270294,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.94801102713576,
+    "estimated_full_shards_per_hour": 64.96537208317449,
     "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": [

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

@@ -232,3 +232,4 @@
 {"step": 338200, "loss": 0.5894011985510588, "lr": 0.0003, "grad_norm": 0.14903607777050254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49503616604, "tokens_per_second": 149798.16330088832, "data_wait_s": 1.1824999091913924e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 338300, "loss": 0.6193603668361902, "lr": 0.0003, "grad_norm": 0.15474905746927944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49518330857, "tokens_per_second": 149156.83811656805, "data_wait_s": 1.237299875356257e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 338400, "loss": 0.5691492799669504, "lr": 0.0003, "grad_norm": 0.14570167071391565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49533027987, "tokens_per_second": 149911.24673057065, "data_wait_s": 8.025999704841524e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 338500, "loss": 0.5700047565624118, "lr": 0.0003, "grad_norm": 0.1598025247832868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49547452233, "tokens_per_second": 147611.6312319545, "data_wait_s": 1.3952001609141007e-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:c1b97ae471c5c19e9cdfdd6fa28d5560edd72c0ae6a0fc3c9db350eeab66b4cc
-size 498909
+oid sha256:5ff457071487ca9f471c2fb63cbf29184d30d7dcb2f14f6331e86465deab2aed
+size 501027