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": 75606.22224577566,
-      "required_pages_per_second_upper_estimate": 0.006553355722045447
+      "required_token_positions_per_second": 75987.53117783446,
+      "required_pages_per_second_upper_estimate": 0.006586406613857652
     },
     {
       "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": 9450.777780721957,
-      "required_pages_per_second_upper_estimate": 0.0037700565584497994
+      "required_token_positions_per_second": 9498.441397229308,
+      "required_pages_per_second_upper_estimate": 0.003789070287709154
     },
     {
       "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": 18901.555561443914,
-      "required_pages_per_second_upper_estimate": 0.004372542223123782
+      "required_token_positions_per_second": 18996.882794458616,
+      "required_pages_per_second_upper_estimate": 0.004394594500779733
     },
     {
       "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": 28352.33334216587,
-      "required_pages_per_second_upper_estimate": 0.001726508347341108
+      "required_token_positions_per_second": 28495.324191687927,
+      "required_pages_per_second_upper_estimate": 0.0017352157398619935
     },
     {
       "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": 28352.33334216587,
-      "required_pages_per_second_upper_estimate": 0.0021826411049618373
+      "required_token_positions_per_second": 28495.324191687927,
+      "required_pages_per_second_upper_estimate": 0.0021936489363820508
     },
     {
       "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": 28352.33334216587,
-      "required_pages_per_second_upper_estimate": 0.3520410909540444
+      "required_token_positions_per_second": 28495.324191687927,
+      "required_pages_per_second_upper_estimate": 0.35381655874551976
     }
   ],
   "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": 0.46268520113275713,
-    "producer_p95_seconds": 0.31840511499831337,
-    "gpu_update_seconds": 0.9751578350301638,
-    "producer_batches_per_second": 2.1612967035724844,
-    "gpu_batches_per_second": 1.0254750196095874,
-    "producer_headroom": 2.107605414313574,
+    "producer_batch_seconds": 0.08864769213275281,
+    "producer_p95_seconds": 0.3067721360021096,
+    "gpu_update_seconds": 0.9702644480902207,
+    "producer_batches_per_second": 11.280609522270105,
+    "gpu_batches_per_second": 1.0306468529979718,
+    "producer_headroom": 10.945174372246692,
     "sustainable": true,
-    "buffer_seconds": 60.45978577187016,
+    "buffer_seconds": 60.15639578159369,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295025.4307403377,
+    "generation_bytes_per_second": 305651.1960948202,
     "upload_bytes_per_second": 5926733.943786841,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.30558959634398,
+    "estimated_full_shards_per_hour": 65.58563148625808,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -171,3 +171,4 @@
 {"step": 132100, "loss": 0.6819533929228783, "lr": 0.0003, "grad_norm": 0.1560970733064769, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19331878226, "tokens_per_second": 148650.67245640198, "data_wait_s": 8.839000656735152e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 132200, "loss": 0.5880027711391449, "lr": 0.0003, "grad_norm": 0.12591463884297227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19346583756, "tokens_per_second": 136815.05019313606, "data_wait_s": 1.4776996977161616e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 132300, "loss": 0.6078644264489412, "lr": 0.0003, "grad_norm": 0.13996294146397065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19361387048, "tokens_per_second": 151207.65245956017, "data_wait_s": 1.1065996659453958e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 132400, "loss": 0.5686908140778542, "lr": 0.0003, "grad_norm": 0.14668115245836916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19375992526, "tokens_per_second": 149281.23510127407, "data_wait_s": 1.4981000276748091e-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 sha256:761573add1e3a148b5f8567f34a1c28acd8b8127320856847b91d42eb0db1736
-size 368928
+oid sha256:ac436725bcefb7fdea2e68982ae09327169fc67efc4a7b3bd9e82bed3b416fff
+size 371046