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": 73373.69615757691,
-      "required_pages_per_second_upper_estimate": 0.0063598460190060204
+      "required_token_positions_per_second": 73080.5461657686,
+      "required_pages_per_second_upper_estimate": 0.006334436520697937
     },
     {
       "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": 9171.712019697114,
-      "required_pages_per_second_upper_estimate": 0.0036587330539720416
+      "required_token_positions_per_second": 9135.068270721074,
+      "required_pages_per_second_upper_estimate": 0.003644115314632629
     },
     {
       "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": 18343.424039394227,
-      "required_pages_per_second_upper_estimate": 0.004243428318276885
+      "required_token_positions_per_second": 18270.13654144215,
+      "required_pages_per_second_upper_estimate": 0.004226474545441583
     },
     {
       "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": 27515.13605909134,
-      "required_pages_per_second_upper_estimate": 0.001675527425236548
+      "required_token_positions_per_second": 27405.20481216322,
+      "required_pages_per_second_upper_estimate": 0.0016688331890633036
     },
     {
       "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": 27515.13605909134,
-      "required_pages_per_second_upper_estimate": 0.002118191340600353
+      "required_token_positions_per_second": 27405.20481216322,
+      "required_pages_per_second_upper_estimate": 0.0021097285289026586
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 27515.13605909134,
-      "required_pages_per_second_upper_estimate": 0.34164590261732297
+      "required_token_positions_per_second": 27405.20481216322,
+      "required_pages_per_second_upper_estimate": 0.34028092444669183
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "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,
+    "producer_batch_seconds": 0.4315791902343733,
+    "producer_p95_seconds": 0.18856492600025376,
+    "gpu_update_seconds": 1.008859455329778,
+    "producer_batches_per_second": 2.3170718668268973,
+    "gpu_batches_per_second": 0.9912183453473388,
+    "producer_headroom": 2.3375998615269356,
     "sustainable": true,
-    "buffer_seconds": 60.289725496446735,
+    "buffer_seconds": 60.53156731978668,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361539.18844424014,
-    "upload_bytes_per_second": 5108206.667733794,
+    "generation_bytes_per_second": 330661.13703039085,
+    "upload_bytes_per_second": 5103747.059018985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.577893638567,
+    "estimated_full_shards_per_hour": 70.95218260940355,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 1134403,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.2444776140000613,
+      "bytes": 6243354,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -52,3 +52,4 @@
 {"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}
+{"step": 200500, "loss": 0.5637441445142031, "lr": 0.0003, "grad_norm": 0.15372752332988915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29345458656, "tokens_per_second": 148831.16892741164, "data_wait_s": 1.547899955767207e-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 sha256:54a10e980e078f9e29828095a590e4c0bc4ecd23e5fbe6ee226a5075282b54f1
-size 115320
+oid sha256:ff458df597f70fbc3b9a843c98719faf9bf5b9ad35b3cffb48f2603b155fb6e5
+size 117438