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": 74812.1529462261,
-      "required_pages_per_second_upper_estimate": 0.006484527807710732
+      "required_token_positions_per_second": 75438.42309869369,
+      "required_pages_per_second_upper_estimate": 0.0065388113172594325
     },
     {
       "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": 9351.519118278262,
-      "required_pages_per_second_upper_estimate": 0.003730460793951756
+      "required_token_positions_per_second": 9429.802887336711,
+      "required_pages_per_second_upper_estimate": 0.003761689359875822
     },
     {
       "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": 18703.038236556524,
-      "required_pages_per_second_upper_estimate": 0.004326618733796674
+      "required_token_positions_per_second": 18859.605774673422,
+      "required_pages_per_second_upper_estimate": 0.00436283787824546
     },
     {
       "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": 28054.557354834786,
-      "required_pages_per_second_upper_estimate": 0.0017083753520225088
+      "required_token_positions_per_second": 28289.408662010133,
+      "required_pages_per_second_upper_estimate": 0.0017226765644599057
     },
     {
       "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": 28054.557354834786,
-      "required_pages_per_second_upper_estimate": 0.0021597174851604004
+      "required_token_positions_per_second": 28289.408662010133,
+      "required_pages_per_second_upper_estimate": 0.0021777969888967857
     },
     {
       "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": 28054.557354834786,
-      "required_pages_per_second_upper_estimate": 0.348343709783513
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28289.408662010133,
+      "required_pages_per_second_upper_estimate": 0.3512597770218674
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.13610226025014072,
-    "producer_p95_seconds": 0.314272802999767,
-    "gpu_update_seconds": 0.985508331153023,
-    "producer_batches_per_second": 7.347416554009551,
-    "gpu_batches_per_second": 1.0147047654381796,
-    "producer_headroom": 7.240940226428048,
+    "producer_batch_seconds": 0.7036411262187414,
+    "producer_p95_seconds": 0.3204896319984982,
+    "gpu_update_seconds": 0.9773268975087669,
+    "producer_batches_per_second": 1.4211790112011293,
+    "gpu_batches_per_second": 1.0231990980183063,
+    "producer_headroom": 1.3889564738217768,
     "sustainable": true,
-    "buffer_seconds": 60.116008200334406,
+    "buffer_seconds": 60.594267645543546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386975.686773351,
-    "upload_bytes_per_second": 4087435.3467923487,
+    "generation_bytes_per_second": 339297.2004835012,
+    "upload_bytes_per_second": 4143614.3058857257,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.03597404861829,
+    "estimated_full_shards_per_hour": 72.80528078917291,
     "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": 2.6395804720013984,
-      "bytes": 7630392,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8558966809978301,
       "bytes": 4770134,
@@ -562,6 +556,12 @@
       "bytes": 7370627,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 0.999219997000182,
+      "bytes": 6670501,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -279,3 +279,4 @@
 {"step": 342900, "loss": 0.6649965941905975, "lr": 0.0003, "grad_norm": 0.1878389021126051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50191531985, "tokens_per_second": 147458.1379250797, "data_wait_s": 1.1872001778101549e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 343000, "loss": 0.5789658762514591, "lr": 0.0003, "grad_norm": 0.16937765798034796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50206134439, "tokens_per_second": 148222.5559338157, "data_wait_s": 1.3883000065106899e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 343100, "loss": 0.5893463902175426, "lr": 0.0003, "grad_norm": 0.16156630115794485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50220664206, "tokens_per_second": 133755.84668161237, "data_wait_s": 1.5747998986626044e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 343200, "loss": 0.5764463171362877, "lr": 0.0003, "grad_norm": 0.13561009381328162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50235274808, "tokens_per_second": 148330.61295416395, "data_wait_s": 1.5412002539960667e-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:a7778345a5a8aaab0c6a87699beed2faeeb742a129322ba386e4d2a4b4d4b0e9
-size 599238
+oid sha256:233ea76d63722b0ee4ff3dee2dc6d31de301325bbb7756bcd5cd3a45f47b1948
+size 601356