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": 74070.13649864245,
-      "required_pages_per_second_upper_estimate": 0.006420211702657674
+      "required_token_positions_per_second": 75208.29364608426,
+      "required_pages_per_second_upper_estimate": 0.006518864279565013
     },
     {
       "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": 9258.767062330306,
-      "required_pages_per_second_upper_estimate": 0.003693460612067299
+      "required_token_positions_per_second": 9401.036705760533,
+      "required_pages_per_second_upper_estimate": 0.003750214099952343
     },
     {
       "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": 18517.53412466061,
-      "required_pages_per_second_upper_estimate": 0.004283705622270419
+      "required_token_positions_per_second": 18802.073411521065,
+      "required_pages_per_second_upper_estimate": 0.004349528778565172
     },
     {
       "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": 27776.301186990917,
-      "required_pages_per_second_upper_estimate": 0.001691431011298103
+      "required_token_positions_per_second": 28203.110117281598,
+      "required_pages_per_second_upper_estimate": 0.00171742143585941
     },
     {
       "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": 27776.301186990917,
-      "required_pages_per_second_upper_estimate": 0.0021382965550974077
+      "required_token_positions_per_second": 28203.110117281598,
+      "required_pages_per_second_upper_estimate": 0.0021711534880339213
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27776.301186990917,
-      "required_pages_per_second_upper_estimate": 0.3448886994423795
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28203.110117281598,
+      "required_pages_per_second_upper_estimate": 0.3501882379189888
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10509689617197182,
-    "producer_p95_seconds": 0.3456709589991078,
-    "gpu_update_seconds": 0.9953809117302124,
-    "producer_batches_per_second": 9.515028858356418,
-    "gpu_batches_per_second": 1.0046405232563267,
-    "producer_headroom": 9.471078100170093,
+    "producer_batch_seconds": 0.4946608948281721,
+    "producer_p95_seconds": 0.3276067160004459,
+    "gpu_update_seconds": 0.980317414817969,
+    "producer_batches_per_second": 2.021586930471561,
+    "gpu_batches_per_second": 1.020077767552141,
+    "producer_headroom": 1.9817968735096736,
     "sustainable": true,
-    "buffer_seconds": 60.71823561554295,
+    "buffer_seconds": 60.77967971871408,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312462.092878963,
+    "generation_bytes_per_second": 277212.3029144476,
     "upload_bytes_per_second": 3931766.5556723503,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.04709138657252,
+    "estimated_full_shards_per_hour": 59.48330703330108,
     "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

@@ -259,3 +259,4 @@
 {"step": 140900, "loss": 0.6342130172997713, "lr": 0.0003, "grad_norm": 0.15846270861170156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20620749311, "tokens_per_second": 149953.54148665606, "data_wait_s": 1.0063002264359966e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 141000, "loss": 0.6059113256633282, "lr": 0.0003, "grad_norm": 0.17628415937895242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20635479752, "tokens_per_second": 149509.69839055766, "data_wait_s": 1.4297002053353935e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 141100, "loss": 0.590720796957612, "lr": 0.0003, "grad_norm": 0.1466954825012965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20649955142, "tokens_per_second": 148243.05961548552, "data_wait_s": 1.7220998415723443e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 141200, "loss": 0.6340356022119522, "lr": 0.0003, "grad_norm": 0.1577793717491402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20664575783, "tokens_per_second": 133969.28370366272, "data_wait_s": 1.1880001693498343e-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 sha256:8cdf4b7a5054749270165b83979df4cc43894c9ddb9a2264252c6b78628df984
-size 556878
+oid sha256:81eda884e2834f55d7f81b94c91e2a3ad3b6de754874a22699859686d48280f3
+size 558996