Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75963.14630241891,
-      "required_pages_per_second_upper_estimate": 0.005255306719984483
+      "required_token_positions_per_second": 76815.53681738363,
+      "required_pages_per_second_upper_estimate": 0.005314277073628228
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9495.393287802364,
-      "required_pages_per_second_upper_estimate": 0.003975522883243107
+      "required_token_positions_per_second": 9601.942102172954,
+      "required_pages_per_second_upper_estimate": 0.004020132646827816
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18990.78657560473,
-      "required_pages_per_second_upper_estimate": 0.003955280022977731
+      "required_token_positions_per_second": 19203.884204345908,
+      "required_pages_per_second_upper_estimate": 0.003999662639282137
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28486.17986340709,
-      "required_pages_per_second_upper_estimate": 0.0017077029869330966
+      "required_token_positions_per_second": 28805.826306518862,
+      "required_pages_per_second_upper_estimate": 0.001726865303125789
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28486.17986340709,
-      "required_pages_per_second_upper_estimate": 0.0022146751869165262
+      "required_token_positions_per_second": 28805.826306518862,
+      "required_pages_per_second_upper_estimate": 0.0022395262918923514
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28486.17986340709,
-      "required_pages_per_second_upper_estimate": 0.3653901292108502
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28805.826306518862,
+      "required_pages_per_second_upper_estimate": 0.36949021057347725
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.23343508467182517,
-    "producer_p95_seconds": 0.22212341299928084,
-    "gpu_update_seconds": 0.9705759119886832,
-    "producer_batches_per_second": 4.283846198209025,
-    "gpu_batches_per_second": 1.0303161119577218,
-    "producer_headroom": 4.157797930645978,
+    "producer_batch_seconds": 0.1690959470860065,
+    "producer_p95_seconds": 0.21223722399918188,
+    "gpu_update_seconds": 0.9598058290639334,
+    "producer_batches_per_second": 5.913802295281356,
+    "gpu_batches_per_second": 1.0418773982392528,
+    "producer_headroom": 5.6761019149427145,
     "sustainable": true,
-    "buffer_seconds": 60.17570654329836,
+    "buffer_seconds": 60.46776723102781,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384329.9433093682,
+    "generation_bytes_per_second": 439410.19464992,
     "upload_bytes_per_second": 5539857.077252538,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.46825909100328,
+    "estimated_full_shards_per_hour": 94.28719882605743,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 617800, "loss": 0.558008074760437, "lr": 0.0001, "grad_norm": 0.17801642857057393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90428248472, "tokens_per_second": 142485.59363283866, "data_wait_s": 1.4054001439944841e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 617900, "loss": 0.5300293564796448, "lr": 0.0001, "grad_norm": 0.16156128265384173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90443108316, "tokens_per_second": 155072.08380046787, "data_wait_s": 1.4151999494060874e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 618000, "loss": 0.5506353974342346, "lr": 0.0001, "grad_norm": 0.1530333902383697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90457754751, "tokens_per_second": 152628.27553322975, "data_wait_s": 1.4923998605809174e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 618100, "loss": 0.5441434383392334, "lr": 0.0001, "grad_norm": 0.16083786768726563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90472583856, "tokens_per_second": 154745.39214966018, "data_wait_s": 1.2869999409304e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef5cfeab6b18d999fb18dde6edf205800dd79812c2b98d0174c577d24f8edbf0
-size 290562
+oid sha256:ef3e5bcab5adf3aea523d41865469534296539405c81a8b293671b7ca1d58c28
+size 292680