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": 76228.668914566,
-      "required_pages_per_second_upper_estimate": 0.005273676190390161
+      "required_token_positions_per_second": 77432.78975251346,
+      "required_pages_per_second_upper_estimate": 0.00535698006390464
     },
     {
       "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": 9528.58361432075,
-      "required_pages_per_second_upper_estimate": 0.0039894189798635234
+      "required_token_positions_per_second": 9679.098719064183,
+      "required_pages_per_second_upper_estimate": 0.004052436511106796
     },
     {
       "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": 19057.1672286415,
-      "required_pages_per_second_upper_estimate": 0.00396910536242975
+      "required_token_positions_per_second": 19358.197438128365,
+      "required_pages_per_second_upper_estimate": 0.004031802016365385
     },
     {
       "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": 28585.750842962254,
-      "required_pages_per_second_upper_estimate": 0.0017136721151213461
+      "required_token_positions_per_second": 29037.296157192548,
+      "required_pages_per_second_upper_estimate": 0.0017407415672396791
     },
     {
       "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": 28585.750842962254,
-      "required_pages_per_second_upper_estimate": 0.0022224163926104867
+      "required_token_positions_per_second": 29037.296157192548,
+      "required_pages_per_second_upper_estimate": 0.0022575220546539564
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28585.750842962254,
-      "required_pages_per_second_upper_estimate": 0.36666731882559556
+      "required_token_positions_per_second": 29037.296157192548,
+      "required_pages_per_second_upper_estimate": 0.3724592572849572
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.40795902450020094,
-    "producer_p95_seconds": 0.32263869399321266,
-    "gpu_update_seconds": 0.9671951648877845,
-    "producer_batches_per_second": 2.451226569200475,
-    "gpu_batches_per_second": 1.0339174928733454,
-    "producer_headroom": 2.3708144857751714,
+    "producer_batch_seconds": 0.11054551308637883,
+    "producer_p95_seconds": 0.336058282002341,
+    "gpu_update_seconds": 0.9521547684856182,
+    "producer_batches_per_second": 9.046047841115115,
+    "gpu_batches_per_second": 1.0502494269817906,
+    "producer_headroom": 8.61323758786679,
     "sustainable": true,
-    "buffer_seconds": 60.93329538793042,
+    "buffer_seconds": 60.937905183079565,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329716.1421738058,
-    "upload_bytes_per_second": 4878668.169995524,
+    "generation_bytes_per_second": 316357.13649097737,
+    "upload_bytes_per_second": 4825419.580442023,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.74940871153478,
+    "estimated_full_shards_per_hour": 67.8828770737361,
     "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": 1.381297531999735,
-      "bytes": 11693702,
-      "pages": 44,
-      "shards": 3
-    },
     {
       "seconds": 1.4175854469976912,
       "bytes": 7625416,
@@ -562,6 +556,12 @@
       "bytes": 7481150,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 2.095004743998288,
+      "bytes": 10399963,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -367,3 +367,4 @@
 {"step": 796700, "loss": 0.5909894108772278, "lr": 3e-05, "grad_norm": 0.28662822347652184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116614287455, "tokens_per_second": 150561.0224878996, "data_wait_s": 1.1718999303411692e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 796800, "loss": 0.5350161790847778, "lr": 3e-05, "grad_norm": 0.26982433464996836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116628817201, "tokens_per_second": 150991.49962865576, "data_wait_s": 1.1568998161237687e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 796900, "loss": 0.48557382822036743, "lr": 3e-05, "grad_norm": 0.26081033136248116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116643322906, "tokens_per_second": 151187.62686386445, "data_wait_s": 1.1613999959081411e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 797000, "loss": 0.5508695244789124, "lr": 3e-05, "grad_norm": 0.2761141875675312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116657996194, "tokens_per_second": 152593.4678277973, "data_wait_s": 1.3636999938171357e-05, "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 sha256:3ebad3bf84aa1fb4f47675c48db9ca6af564a5230f2d2faf03384dabd87c5ac3
-size 787188
+oid sha256:4a38c034ebeeb7e9477735371e893c35116066d7c7725ed823b71ca81d65c5a6
+size 789306