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": 75821.11651308216,
-      "required_pages_per_second_upper_estimate": 0.006571982212491034
+      "required_token_positions_per_second": 75799.34966219935,
+      "required_pages_per_second_upper_estimate": 0.006570095517024092
     },
     {
       "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": 9477.63956413527,
-      "required_pages_per_second_upper_estimate": 0.003780772125472822
+      "required_token_positions_per_second": 9474.91870777492,
+      "required_pages_per_second_upper_estimate": 0.0037796867351902502
     },
     {
       "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": 18955.27912827054,
-      "required_pages_per_second_upper_estimate": 0.004384970224806639
+      "required_token_positions_per_second": 18949.83741554984,
+      "required_pages_per_second_upper_estimate": 0.004383711380339577
     },
     {
       "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": 28432.91869240581,
-      "required_pages_per_second_upper_estimate": 0.001731415572372063
+      "required_token_positions_per_second": 28424.756123324758,
+      "required_pages_per_second_upper_estimate": 0.0017309185147407174
     },
     {
       "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": 28432.91869240581,
-      "required_pages_per_second_upper_estimate": 0.0021888447882978347
+      "required_token_positions_per_second": 28424.756123324758,
+      "required_pages_per_second_upper_estimate": 0.002188216411134538
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28432.91869240581,
-      "required_pages_per_second_upper_estimate": 0.35304169130220653
+      "required_token_positions_per_second": 28424.756123324758,
+      "required_pages_per_second_upper_estimate": 0.35294033951258125
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11330460203873827,
-    "producer_p95_seconds": 0.376073257997632,
-    "gpu_update_seconds": 0.97239401621419,
-    "producer_batches_per_second": 8.825766844475611,
-    "gpu_batches_per_second": 1.0283897096500945,
-    "producer_headroom": 8.582122868069678,
+    "producer_batch_seconds": 0.2785964577503819,
+    "producer_p95_seconds": 0.39459214900125517,
+    "gpu_update_seconds": 0.9726732528520318,
+    "producer_batches_per_second": 3.589421086236439,
+    "gpu_batches_per_second": 1.0280944778401604,
+    "producer_headroom": 3.4913338838052708,
     "sustainable": true,
-    "buffer_seconds": 60.28842900527978,
+    "buffer_seconds": 60.30574167682597,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 551422.2772046723,
+    "generation_bytes_per_second": 552878.6765362287,
     "upload_bytes_per_second": 5780483.223379054,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.32238423447731,
+    "estimated_full_shards_per_hour": 118.63489362778803,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -338,3 +338,4 @@
 {"step": 228800, "loss": 0.5507429484277964, "lr": 0.0003, "grad_norm": 0.14181171220754413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33489760417, "tokens_per_second": 146354.72225646026, "data_wait_s": 9.732000762596726e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 228900, "loss": 0.6200902182608843, "lr": 0.0003, "grad_norm": 0.1488659235463212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33504225219, "tokens_per_second": 147854.12281896625, "data_wait_s": 9.461997251491994e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 229000, "loss": 0.5841774679720402, "lr": 0.0003, "grad_norm": 0.15640982062436973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33518730851, "tokens_per_second": 146847.44436192018, "data_wait_s": 1.1139003618154675e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 229100, "loss": 0.6136814542114735, "lr": 0.0003, "grad_norm": 0.16188222887238107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33533374596, "tokens_per_second": 149203.84136049138, "data_wait_s": 1.109000004362315e-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 sha256:16f8afbb8e1204173f9ad86eddf8d42532fd48c2f83871b87e78c4b77b7adf52
-size 724983
+oid sha256:06c7499d85f4bae54b379a0a937f29710dd6744000a0d94f1882cf42ae0b1c14
+size 727101