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": 75111.69848789749,
-      "required_pages_per_second_upper_estimate": 0.006510491656071574
+      "required_token_positions_per_second": 74702.52149831278,
+      "required_pages_per_second_upper_estimate": 0.00647502523166397
     },
     {
       "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": 9388.962310987186,
-      "required_pages_per_second_upper_estimate": 0.0037453974433489653
+      "required_token_positions_per_second": 9337.815187289098,
+      "required_pages_per_second_upper_estimate": 0.003724994090988151
     },
     {
       "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": 18777.924621974373,
-      "required_pages_per_second_upper_estimate": 0.004343942381107724
+      "required_token_positions_per_second": 18675.630374578195,
+      "required_pages_per_second_upper_estimate": 0.004320278407289884
     },
     {
       "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": 28166.88693296156,
-      "required_pages_per_second_upper_estimate": 0.00171521563398268
+      "required_token_positions_per_second": 28013.44556186729,
+      "required_pages_per_second_upper_estimate": 0.001705871859527696
     },
     {
       "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": 28166.88693296156,
-      "required_pages_per_second_upper_estimate": 0.002168364927032773
+      "required_token_positions_per_second": 28013.44556186729,
+      "required_pages_per_second_upper_estimate": 0.0021565525855330356
     },
     {
       "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": 28166.88693296156,
-      "required_pages_per_second_upper_estimate": 0.3497384672009332
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28013.44556186729,
+      "required_pages_per_second_upper_estimate": 0.3478332389071767
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.06923710451565057,
-    "producer_p95_seconds": 0.24123684300138848,
-    "gpu_update_seconds": 0.9815781227724408,
-    "producer_batches_per_second": 14.443122759039655,
-    "gpu_batches_per_second": 1.0187676118692692,
-    "producer_headroom": 14.17705332479006,
+    "producer_batch_seconds": 0.17935652532815993,
+    "producer_p95_seconds": 0.30972496700269403,
+    "gpu_update_seconds": 0.9869546371559254,
+    "producer_batches_per_second": 5.575487137534297,
+    "gpu_batches_per_second": 1.0132177937596676,
+    "producer_headroom": 5.502752884792691,
     "sustainable": true,
-    "buffer_seconds": 60.85784361189133,
+    "buffer_seconds": 60.20423286651145,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289832.970589768,
-    "upload_bytes_per_second": 5190430.702701169,
+    "generation_bytes_per_second": 325382.84480940684,
+    "upload_bytes_per_second": 5222498.231722886,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.1914085223177,
+    "estimated_full_shards_per_hour": 69.8195839711347,
     "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": 1.0256725109993567,
-      "bytes": 2318063,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9941544879984576,
       "bytes": 6059545,
@@ -562,6 +556,12 @@
       "bytes": 6148950,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2174438740003097,
+      "bytes": 5854390,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -202,3 +202,4 @@
 {"step": 215200, "loss": 0.5364100411534309, "lr": 0.0003, "grad_norm": 0.1523177941082317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31498437528, "tokens_per_second": 148927.4257286477, "data_wait_s": 1.3799999578623101e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 215300, "loss": 0.5852600494399667, "lr": 0.0003, "grad_norm": 0.16907311263407124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31513081962, "tokens_per_second": 148980.0537565134, "data_wait_s": 9.14499833015725e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 215400, "loss": 0.5157265942543745, "lr": 0.0003, "grad_norm": 0.1502006032714767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31527713895, "tokens_per_second": 148299.84229734525, "data_wait_s": 7.127997378120199e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 215500, "loss": 0.6108162812888622, "lr": 0.0003, "grad_norm": 0.16144090286693072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31542436363, "tokens_per_second": 133311.82707342232, "data_wait_s": 1.756699930410832e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9630436a9dce8df85c3edf7a65251615e0714030b2ae515a719ed535cdc4115a
-size 435369
+oid sha256:2d44dd9af1735ba67c110d89be9f901f59cc03107195d099a50218fbfd2b17d4
+size 437487