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": 73737.17173406412,
-      "required_pages_per_second_upper_estimate": 0.006391351160755505
+      "required_token_positions_per_second": 75712.98248089162,
+      "required_pages_per_second_upper_estimate": 0.006562609428643957
     },
     {
       "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": 9217.146466758015,
-      "required_pages_per_second_upper_estimate": 0.0036768575342101544
+      "required_token_positions_per_second": 9464.122810111452,
+      "required_pages_per_second_upper_estimate": 0.0037753800901992388
     },
     {
       "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": 18434.29293351603,
-      "required_pages_per_second_upper_estimate": 0.0042644492376940485
+      "required_token_positions_per_second": 18928.245620222904,
+      "required_pages_per_second_upper_estimate": 0.0043787164984933095
     },
     {
       "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": 27651.439400274045,
-      "required_pages_per_second_upper_estimate": 0.0016838275835862116
+      "required_token_positions_per_second": 28392.368430334358,
+      "required_pages_per_second_upper_estimate": 0.0017289462741626946
     },
     {
       "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": 27651.439400274045,
-      "required_pages_per_second_upper_estimate": 0.002128684349116395
+      "required_token_positions_per_second": 28392.368430334358,
+      "required_pages_per_second_upper_estimate": 0.0021857231168732646
     },
     {
       "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": 27651.439400274045,
-      "required_pages_per_second_upper_estimate": 0.3433383339368743
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28392.368430334358,
+      "required_pages_per_second_upper_estimate": 0.35253819276027615
     }
   ],
   "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": 0.3893461598906338,
-    "producer_p95_seconds": 0.13445906300012211,
-    "gpu_update_seconds": 0.9998756158685175,
-    "producer_batches_per_second": 2.568408534659484,
-    "gpu_batches_per_second": 1.0001243996048192,
-    "producer_headroom": 2.568089065394608,
+    "producer_batch_seconds": 0.09333942985155907,
+    "producer_p95_seconds": 0.1401815029998943,
+    "gpu_update_seconds": 0.9737827989883692,
+    "producer_batches_per_second": 10.713585904588603,
+    "gpu_batches_per_second": 1.0269230479721627,
+    "producer_headroom": 10.432705669372629,
     "sustainable": true,
-    "buffer_seconds": 60.992412567979564,
+    "buffer_seconds": 60.374533537278886,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 55,
+    "upload_pages_per_commit": 56,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272172.2591061277,
-    "upload_bytes_per_second": 4077761.177928621,
+    "generation_bytes_per_second": 318544.2515321579,
+    "upload_bytes_per_second": 4084854.416422639,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.40183095825075,
+    "estimated_full_shards_per_hour": 68.35218104814102,
     "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": [
@@ -208,6 +208,12 @@
       "bytes": 6129878,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.5951903079999283,
+      "bytes": 6553135,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -5,3 +5,4 @@
 {"step": 315500, "loss": 0.5826609097421169, "lr": 0.0003, "grad_norm": 0.14364552481390072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46180736323, "tokens_per_second": 151488.22508508177, "data_wait_s": 9.485999953540158e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 315600, "loss": 0.5520508475601673, "lr": 0.0003, "grad_norm": 0.1444274193655029, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46195213605, "tokens_per_second": 146664.88977618984, "data_wait_s": 1.219699993271206e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 315700, "loss": 0.587482888251543, "lr": 0.0003, "grad_norm": 0.15449887059885634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46209767410, "tokens_per_second": 148647.08791861043, "data_wait_s": 1.1756000048990245e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 315800, "loss": 0.5816597454249859, "lr": 0.0003, "grad_norm": 0.15681656811833933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46224346598, "tokens_per_second": 148903.31554262608, "data_wait_s": 1.0683000027711387e-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 sha256:1f1a12407e60bccc6c3791ae569a9b86b326111493db20ab8bdb27077bf1e30a
-size 14991
+oid sha256:763d20dd93604e52b945c587a5f251f8cb99960caf0c1d5777ffbd59ecefb2ec
+size 17109