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": 75076.94630158915,
-      "required_pages_per_second_upper_estimate": 0.006507479424640978
+      "required_token_positions_per_second": 74737.19776124782,
+      "required_pages_per_second_upper_estimate": 0.006478030882248994
     },
     {
       "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": 9384.618287698644,
-      "required_pages_per_second_upper_estimate": 0.0037436645475102297
+      "required_token_positions_per_second": 9342.149720155978,
+      "required_pages_per_second_upper_estimate": 0.0037267232009557914
     },
     {
       "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": 18769.236575397288,
-      "required_pages_per_second_upper_estimate": 0.004341932554436513
+      "required_token_positions_per_second": 18684.299440311956,
+      "required_pages_per_second_upper_estimate": 0.004322283843077044
     },
     {
       "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": 28153.854863095934,
-      "required_pages_per_second_upper_estimate": 0.0017144220493018499
+      "required_token_positions_per_second": 28026.449160467935,
+      "required_pages_per_second_upper_estimate": 0.0017066637104578661
     },
     {
       "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": 28153.854863095934,
-      "required_pages_per_second_upper_estimate": 0.002167361682219439
+      "required_token_positions_per_second": 28026.449160467935,
+      "required_pages_per_second_upper_estimate": 0.0021575536385495782
     },
     {
       "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": 28153.854863095934,
-      "required_pages_per_second_upper_estimate": 0.3495766525087343
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28026.449160467935,
+      "required_pages_per_second_upper_estimate": 0.34799470008155176
     }
   ],
   "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": 0.12057164901554529,
-    "producer_p95_seconds": 0.31892696900104056,
-    "gpu_update_seconds": 0.9820324830984688,
-    "producer_batches_per_second": 8.293823698729293,
-    "gpu_batches_per_second": 1.018296255175634,
-    "producer_headroom": 8.144804281244054,
+    "producer_batch_seconds": 0.5019025871795293,
+    "producer_p95_seconds": 0.31836666599701857,
+    "gpu_update_seconds": 0.9864967139325752,
+    "producer_batches_per_second": 1.992418500210485,
+    "gpu_batches_per_second": 1.013688120676647,
+    "producer_headroom": 1.9655143032361135,
     "sustainable": true,
-    "buffer_seconds": 60.88601395210507,
+    "buffer_seconds": 60.17629954988709,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436410.35616283654,
-    "upload_bytes_per_second": 5860836.405875324,
+    "generation_bytes_per_second": 359219.77760649566,
+    "upload_bytes_per_second": 5841288.119573328,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.64350331939528,
+    "estimated_full_shards_per_hour": 77.08020206590798,
     "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": 1.5194503539987636,
-      "bytes": 9946744,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.188283532999776,
       "bytes": 2401103,
@@ -562,6 +556,12 @@
       "bytes": 7844213,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0721679589987616,
+      "bytes": 5937253,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -166,3 +166,4 @@
 {"step": 331600, "loss": 0.5989936012774706, "lr": 0.0003, "grad_norm": 0.15591725490170522, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48537507560, "tokens_per_second": 149936.8862259838, "data_wait_s": 1.0395997378509492e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 331700, "loss": 0.5897045414894819, "lr": 0.0003, "grad_norm": 0.16212551854037918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48552237994, "tokens_per_second": 149983.39380913397, "data_wait_s": 1.8827002350008115e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 331800, "loss": 0.5697401706129313, "lr": 0.0003, "grad_norm": 0.14947999205298454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48566944005, "tokens_per_second": 126934.52762617593, "data_wait_s": 1.608900129212998e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 331900, "loss": 0.5924331322312355, "lr": 0.0003, "grad_norm": 0.16922174544390134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48581636135, "tokens_per_second": 148586.97467842538, "data_wait_s": 1.1873002222273499e-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:c91cb21a38ee186c6a2819d1ead013c1d46b522b5e454281c0588f081c0b2e78
-size 358338
+oid sha256:508657b3ec201d2f4bcf570c7f544f20d00b1b93c1905d1c1d97a2742ad771e9
+size 360456