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": 73860.90469939099,
-      "required_pages_per_second_upper_estimate": 0.006402076020591704
+      "required_token_positions_per_second": 72923.36135073398,
+      "required_pages_per_second_upper_estimate": 0.0063208121393120075
     },
     {
       "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": 9232.613087423873,
-      "required_pages_per_second_upper_estimate": 0.0036830274004403516
+      "required_token_positions_per_second": 9115.420168841747,
+      "required_pages_per_second_upper_estimate": 0.0036362773930276636
     },
     {
       "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": 18465.226174847747,
-      "required_pages_per_second_upper_estimate": 0.004271605098669687
+      "required_token_positions_per_second": 18230.840337683494,
+      "required_pages_per_second_upper_estimate": 0.004217384060291584
     },
     {
       "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": 27697.839262271624,
-      "required_pages_per_second_upper_estimate": 0.0016866530917406025
+      "required_token_positions_per_second": 27346.26050652524,
+      "required_pages_per_second_upper_estimate": 0.0016652437901068216
     },
     {
       "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": 27697.839262271624,
-      "required_pages_per_second_upper_estimate": 0.0021322563389360076
+      "required_token_positions_per_second": 27346.26050652524,
+      "required_pages_per_second_upper_estimate": 0.0021051908330863654
     },
     {
       "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": 27697.839262271624,
-      "required_pages_per_second_upper_estimate": 0.34391446493253564
+      "required_token_positions_per_second": 27346.26050652524,
+      "required_pages_per_second_upper_estimate": 0.3395490334445688
     }
   ],
   "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": 61,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2422288751950532,
-    "producer_p95_seconds": 0.4693971189990407,
-    "gpu_update_seconds": 0.9982006082929541,
-    "producer_batches_per_second": 4.128326976685817,
-    "gpu_batches_per_second": 1.001802635354153,
-    "producer_headroom": 4.120898499359995,
+    "producer_batch_seconds": 0.10529878528149084,
+    "producer_p95_seconds": 0.43309841900190804,
+    "gpu_update_seconds": 1.0110340312673742,
+    "producer_batches_per_second": 9.49678571625249,
+    "gpu_batches_per_second": 0.9890863898482798,
+    "producer_headroom": 9.601573546785172,
     "sustainable": true,
-    "buffer_seconds": 60.8902371058702,
+    "buffer_seconds": 60.66204187604245,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293738.4346783641,
+    "generation_bytes_per_second": 293493.73707738944,
     "upload_bytes_per_second": 5493555.94485042,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.02943020117942,
+    "estimated_full_shards_per_hour": 62.97692379227888,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -422,3 +422,4 @@
 {"step": 357200, "loss": 0.6047567324712873, "lr": 0.0003, "grad_norm": 0.15845736428146975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52283949229, "tokens_per_second": 149524.7630702091, "data_wait_s": 1.6414000128861517e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357300, "loss": 0.580042626708746, "lr": 0.0003, "grad_norm": 0.156638364723009, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52298628386, "tokens_per_second": 147576.94735376287, "data_wait_s": 1.0554002074059099e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357400, "loss": 0.6024862378835678, "lr": 0.0003, "grad_norm": 0.17464307222932632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52313238998, "tokens_per_second": 148470.39450399738, "data_wait_s": 8.96799610927701e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 357500, "loss": 0.6064887754619122, "lr": 0.0003, "grad_norm": 0.15418517736130338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52327985290, "tokens_per_second": 147869.70254859925, "data_wait_s": 1.3831006071995944e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8bde3d6347f355278fe26dfb8b9ba1b6706bab4a94e99effff430a9a8e90323e
-size 904461
+oid sha256:4b45017d561eded48e97476b498770c738e67e2645e32b870e3f3f85a2742223
+size 906579