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": 74927.80329954586,
-      "required_pages_per_second_upper_estimate": 0.006494552087223348
+      "required_token_positions_per_second": 75025.3383394468,
+      "required_pages_per_second_upper_estimate": 0.006503006177281663
     },
     {
       "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": 9365.975412443233,
-      "required_pages_per_second_upper_estimate": 0.0037362276258350218
+      "required_token_positions_per_second": 9378.16729243085,
+      "required_pages_per_second_upper_estimate": 0.003741091149046932
     },
     {
       "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": 18731.950824886466,
-      "required_pages_per_second_upper_estimate": 0.004333307152262631
+      "required_token_positions_per_second": 18756.3345848617,
+      "required_pages_per_second_upper_estimate": 0.004338947905993383
     },
     {
       "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": 28097.9262373297,
-      "required_pages_per_second_upper_estimate": 0.0017110162894275077
+      "required_token_positions_per_second": 28134.50187729255,
+      "required_pages_per_second_upper_estimate": 0.001713243554003693
     },
     {
       "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": 28097.9262373297,
-      "required_pages_per_second_upper_estimate": 0.002163056141787609
+      "required_token_positions_per_second": 28134.50187729255,
+      "required_pages_per_second_upper_estimate": 0.002165871835799801
     },
     {
       "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": 28097.9262373297,
-      "required_pages_per_second_upper_estimate": 0.3488822061577871
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28134.50187729255,
+      "required_pages_per_second_upper_estimate": 0.3493363531953332
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10628275486709526,
-    "producer_p95_seconds": 0.28393505399799324,
-    "gpu_update_seconds": 0.9839872083964707,
-    "producer_batches_per_second": 9.408864130878829,
-    "gpu_batches_per_second": 1.0162733737460106,
-    "producer_headroom": 9.258201950325144,
+    "producer_batch_seconds": 0.09676556005484827,
+    "producer_p95_seconds": 0.32951248199969996,
+    "gpu_update_seconds": 0.9827079974824361,
+    "producer_batches_per_second": 10.334255280837356,
+    "gpu_batches_per_second": 1.0175962773905003,
+    "producer_headroom": 10.155555312503969,
     "sustainable": true,
-    "buffer_seconds": 60.02321971218471,
+    "buffer_seconds": 60.92789584391104,
     "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": 432060.3123102381,
+    "generation_bytes_per_second": 429313.08519438014,
     "upload_bytes_per_second": 4054113.2021912276,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.71008517246587,
+    "estimated_full_shards_per_hour": 92.12059418557695,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -204,3 +204,4 @@
 {"step": 105400, "loss": 0.6506398729979992, "lr": 0.0003, "grad_norm": 0.158062071480736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15423942472, "tokens_per_second": 147145.7724573228, "data_wait_s": 1.4568999176844954e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 105500, "loss": 0.6525284089148045, "lr": 0.0003, "grad_norm": 0.13751431348209442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15438404912, "tokens_per_second": 146122.16865577575, "data_wait_s": 1.992599936784245e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 105600, "loss": 0.6261213459074497, "lr": 0.0003, "grad_norm": 0.13807369371562536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15453003192, "tokens_per_second": 147776.1816368568, "data_wait_s": 1.63670010806527e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 105700, "loss": 0.6073123849928379, "lr": 0.0003, "grad_norm": 0.1396080629189812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15467619516, "tokens_per_second": 148694.89898032256, "data_wait_s": 1.371300095343031e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ab5e3c29569ed2318172b155b9c80d8dc1e7fdc628b21b0d18c38e0f63d7af3f
-size 439605
+oid sha256:8a7a61bd878733c39965dcfe41633913806568b6e21e0021b75b3500ad779303
+size 441723