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": 75613.86402887992,
-      "required_pages_per_second_upper_estimate": 0.006554018092437003
+      "required_token_positions_per_second": 74442.15355508172,
+      "required_pages_per_second_upper_estimate": 0.00645245719824122
     },
     {
       "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": 9451.73300360999,
-      "required_pages_per_second_upper_estimate": 0.003770437611141691
+      "required_token_positions_per_second": 9305.269194385215,
+      "required_pages_per_second_upper_estimate": 0.0037120110078128353
     },
     {
       "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": 18903.46600721998,
-      "required_pages_per_second_upper_estimate": 0.004372984171131377
+      "required_token_positions_per_second": 18610.53838877043,
+      "required_pages_per_second_upper_estimate": 0.004305220521953082
     },
     {
       "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": 28355.199010829972,
-      "required_pages_per_second_upper_estimate": 0.0017266828515277502
+      "required_token_positions_per_second": 27915.807583155645,
+      "required_pages_per_second_upper_estimate": 0.0016999262188910412
     },
     {
       "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": 28355.199010829972,
-      "required_pages_per_second_upper_estimate": 0.0021828617120682784
+      "required_token_positions_per_second": 27915.807583155645,
+      "required_pages_per_second_upper_estimate": 0.0021490361436525894
     },
     {
       "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": 28355.199010829972,
-      "required_pages_per_second_upper_estimate": 0.35207667296807643
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27915.807583155645,
+      "required_pages_per_second_upper_estimate": 0.3466209019848721
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07870127964845608,
-    "producer_p95_seconds": 0.17468036499940354,
-    "gpu_update_seconds": 0.9750592824067338,
-    "producer_batches_per_second": 12.706273703131808,
-    "gpu_batches_per_second": 1.0255786679264314,
-    "producer_headroom": 12.389370119039253,
+    "producer_batch_seconds": 0.4919447023671637,
+    "producer_p95_seconds": 0.19620616200063523,
+    "gpu_update_seconds": 0.9904065973245482,
+    "producer_batches_per_second": 2.032748793082131,
+    "gpu_batches_per_second": 1.0096863275157568,
+    "producer_headroom": 2.013247815372055,
     "sustainable": true,
-    "buffer_seconds": 60.4536755092175,
+    "buffer_seconds": 60.41480243679744,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375426.00661909964,
-    "upload_bytes_per_second": 5579296.814226379,
+    "generation_bytes_per_second": 343403.97131594695,
+    "upload_bytes_per_second": 5237575.901051338,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.55768155030958,
+    "estimated_full_shards_per_hour": 73.68649820908362,
     "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": [
@@ -226,6 +226,12 @@
       "bytes": 6140767,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.0060759970001527,
+      "bytes": 2526019,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -56,3 +56,4 @@
 {"step": 120600, "loss": 0.5935172829777002, "lr": 0.0003, "grad_norm": 0.13671291106910105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17648295800, "tokens_per_second": 150081.69821159675, "data_wait_s": 9.203999979945365e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 120700, "loss": 0.6122833862900734, "lr": 0.0003, "grad_norm": 0.14585721939975288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17663007237, "tokens_per_second": 150323.7249672456, "data_wait_s": 1.301300017075846e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 120800, "loss": 0.5996634922921658, "lr": 0.0003, "grad_norm": 0.14145665261984264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17677613798, "tokens_per_second": 141081.09919191344, "data_wait_s": 1.4481000107480213e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 120900, "loss": 0.6196673363447189, "lr": 0.0003, "grad_norm": 0.14226493327640644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17692295129, "tokens_per_second": 148894.11960489125, "data_wait_s": 1.4530000044032931e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bab85ad422c9129837e2390b1aac5aa02a5500e91e266cb721d624f795e14cdf
-size 123792
+oid sha256:5d8e6b9d7862aea16b556018bd1ec7b94760457b711ecc52125f1f4ab3867b94
+size 125910