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": 75101.2085490877,
-      "required_pages_per_second_upper_estimate": 0.006509582414762048
+      "required_token_positions_per_second": 74688.9779254429,
+      "required_pages_per_second_upper_estimate": 0.00647385130909348
     },
     {
       "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": 9387.651068635963,
-      "required_pages_per_second_upper_estimate": 0.003744874369170242
+      "required_token_positions_per_second": 9336.122240680363,
+      "required_pages_per_second_upper_estimate": 0.0037243187492741196
     },
     {
       "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": 18775.302137271927,
-      "required_pages_per_second_upper_estimate": 0.004343335715425968
+      "required_token_positions_per_second": 18672.244481360725,
+      "required_pages_per_second_upper_estimate": 0.0043194951404837085
     },
     {
       "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": 28162.95320590789,
-      "required_pages_per_second_upper_estimate": 0.0017149760906437822
+      "required_token_positions_per_second": 28008.36672204109,
+      "required_pages_per_second_upper_estimate": 0.0017055625848288937
     },
     {
       "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": 28162.95320590789,
-      "required_pages_per_second_upper_estimate": 0.002168062097834928
+      "required_token_positions_per_second": 28008.36672204109,
+      "required_pages_per_second_upper_estimate": 0.0021561616023840864
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28162.95320590789,
-      "required_pages_per_second_upper_estimate": 0.3496896234762642
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28008.36672204109,
+      "required_pages_per_second_upper_estimate": 0.34777017671431876
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.3541572145625196,
-    "producer_p95_seconds": 0.35957608399985475,
-    "gpu_update_seconds": 0.9817152270167511,
-    "producer_batches_per_second": 2.8236047689590955,
-    "gpu_batches_per_second": 1.018625332968312,
-    "producer_headroom": 2.7719757967642598,
+    "producer_batch_seconds": 0.11468796975796636,
+    "producer_p95_seconds": 0.35016443299900857,
+    "gpu_update_seconds": 0.9871336045540456,
+    "producer_batches_per_second": 8.719310334905801,
+    "gpu_batches_per_second": 1.0130340972960463,
+    "producer_headroom": 8.607124240120907,
     "sustainable": true,
-    "buffer_seconds": 60.86634407503857,
+    "buffer_seconds": 60.21514987779678,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335126.0417429608,
-    "upload_bytes_per_second": 4146479.810670086,
+    "generation_bytes_per_second": 320235.1327228681,
+    "upload_bytes_per_second": 4138510.307899638,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.91024722305887,
+    "estimated_full_shards_per_hour": 68.71500478996785,
     "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": 1.0725273009993543,
-      "bytes": 11023216,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 2.0963671690005867,
       "bytes": 7398466,
@@ -562,6 +556,12 @@
       "bytes": 1045811,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.2339288350012794,
+      "bytes": 10876745,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -290,3 +290,4 @@
 {"step": 344000, "loss": 0.5531006511300802, "lr": 0.0003, "grad_norm": 0.14453123897165288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50352088987, "tokens_per_second": 148383.51602231924, "data_wait_s": 1.5735000488348305e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 344100, "loss": 0.5920410379767418, "lr": 0.0003, "grad_norm": 0.16994119213599157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50366769807, "tokens_per_second": 149389.57392022002, "data_wait_s": 1.4823999663349241e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 344200, "loss": 0.5796448960900307, "lr": 0.0003, "grad_norm": 0.15931724983149692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50381281804, "tokens_per_second": 146848.80195492788, "data_wait_s": 1.6038997273426503e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 344300, "loss": 0.6537228301167488, "lr": 0.0003, "grad_norm": 0.18833676195820212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50395982680, "tokens_per_second": 149616.36619675183, "data_wait_s": 1.4023000403540209e-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:d8ad6aa940b0034594f7f9fffc30a5d5cea5c9b1f8073000d826a097a195c27c
-size 622536
+oid sha256:215950ba74bb02b706322ca34a1bd36e6db738ceb27f250f36d880b870f8f949
+size 624654