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": 75105.79589872155,
-      "required_pages_per_second_upper_estimate": 0.006509980034601779
+      "required_token_positions_per_second": 73373.69615757691,
+      "required_pages_per_second_upper_estimate": 0.0063598460190060204
     },
     {
       "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": 9388.224487340194,
-      "required_pages_per_second_upper_estimate": 0.0037451031144647338
+      "required_token_positions_per_second": 9171.712019697114,
+      "required_pages_per_second_upper_estimate": 0.0036587330539720416
     },
     {
       "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": 18776.44897468039,
-      "required_pages_per_second_upper_estimate": 0.0043436010160767665
+      "required_token_positions_per_second": 18343.424039394227,
+      "required_pages_per_second_upper_estimate": 0.004243428318276885
     },
     {
       "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": 28164.673462020583,
-      "required_pages_per_second_upper_estimate": 0.0017150808452155591
+      "required_token_positions_per_second": 27515.13605909134,
+      "required_pages_per_second_upper_estimate": 0.001675527425236548
     },
     {
       "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": 28164.673462020583,
-      "required_pages_per_second_upper_estimate": 0.002168194527912457
+      "required_token_positions_per_second": 27515.13605909134,
+      "required_pages_per_second_upper_estimate": 0.002118191340600353
     },
     {
       "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": 28164.673462020583,
-      "required_pages_per_second_upper_estimate": 0.349710983299857
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27515.13605909134,
+      "required_pages_per_second_upper_estimate": 0.34164590261732297
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10973747467195238,
-    "producer_p95_seconds": 0.20396784500007925,
-    "gpu_update_seconds": 0.9816552653196102,
-    "producer_batches_per_second": 9.112657303162711,
-    "gpu_batches_per_second": 1.0186875528797954,
-    "producer_headroom": 8.945488022702875,
+    "producer_batch_seconds": 0.07460495116406918,
+    "producer_p95_seconds": 0.18155038399982004,
+    "gpu_update_seconds": 1.0048287582741122,
+    "producer_batches_per_second": 13.403936124840122,
+    "gpu_batches_per_second": 0.995194446581718,
+    "producer_headroom": 13.468660492308617,
     "sustainable": true,
-    "buffer_seconds": 60.862626449815835,
+    "buffer_seconds": 60.289725496446735,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361227.16107171203,
-    "upload_bytes_per_second": 5265772.158304157,
+    "generation_bytes_per_second": 361539.18844424014,
+    "upload_bytes_per_second": 5108206.667733794,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.51093982804795,
+    "estimated_full_shards_per_hour": 77.577893638567,
     "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": [
@@ -292,6 +292,12 @@
       "bytes": 7654224,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.9410484550007823,
+      "bytes": 1134403,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -51,3 +51,4 @@
 {"step": 200100, "loss": 0.7128876578062773, "lr": 0.0003, "grad_norm": 0.224431648026927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29286561363, "tokens_per_second": 136632.69797399724, "data_wait_s": 2.6620999960869085e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 200200, "loss": 0.5220747478306293, "lr": 0.0003, "grad_norm": 0.18768785979315505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29301246852, "tokens_per_second": 147670.59257115802, "data_wait_s": 1.62049991558888e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 200300, "loss": 0.571290036663413, "lr": 0.0003, "grad_norm": 0.14327233591459235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29315903327, "tokens_per_second": 147597.73080269733, "data_wait_s": 1.884399989648955e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 200400, "loss": 0.591425783932209, "lr": 0.0003, "grad_norm": 0.1432030194790905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29330744063, "tokens_per_second": 126901.42199040274, "data_wait_s": 1.7641000340518076e-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 sha256:62507838d87446e839c2faf958260943db5eb96c2ee3fea32a41b1ae079154ba
-size 113202
+oid sha256:54a10e980e078f9e29828095a590e4c0bc4ecd23e5fbe6ee226a5075282b54f1
+size 115320