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": 74504.48397394913,
-      "required_pages_per_second_upper_estimate": 0.006457859840973646
+      "required_token_positions_per_second": 74209.06777332278,
+      "required_pages_per_second_upper_estimate": 0.006432253913428869
     },
     {
       "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": 9313.060496743641,
-      "required_pages_per_second_upper_estimate": 0.0037151190748139626
+      "required_token_positions_per_second": 9276.133471665347,
+      "required_pages_per_second_upper_estimate": 0.0037003883324020054
     },
     {
       "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": 18626.120993487282,
-      "required_pages_per_second_upper_estimate": 0.004308825283309849
+      "required_token_positions_per_second": 18552.266943330695,
+      "required_pages_per_second_upper_estimate": 0.004291740448593012
     },
     {
       "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": 27939.18149023092,
-      "required_pages_per_second_upper_estimate": 0.0017013495671985126
+      "required_token_positions_per_second": 27828.400414996042,
+      "required_pages_per_second_upper_estimate": 0.001694603580940091
     },
     {
       "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": 27939.18149023092,
-      "required_pages_per_second_upper_estimate": 0.00215083553172236
+      "required_token_positions_per_second": 27828.400414996042,
+      "required_pages_per_second_upper_estimate": 0.0021423072978891356
     },
     {
       "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": 27939.18149023092,
-      "required_pages_per_second_upper_estimate": 0.3469111276833123
+      "required_token_positions_per_second": 27828.400414996042,
+      "required_pages_per_second_upper_estimate": 0.34553559748930357
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13786269940641205,
-    "producer_p95_seconds": 0.2844707260010182,
-    "gpu_update_seconds": 0.9895780235962627,
-    "producer_batches_per_second": 7.25359364284644,
-    "gpu_batches_per_second": 1.0105317379279124,
-    "producer_headroom": 7.177996861058395,
+    "producer_batch_seconds": 0.15247885282835227,
+    "producer_p95_seconds": 0.28426007899906836,
+    "gpu_update_seconds": 0.9935173990489649,
+    "producer_batches_per_second": 6.558286486623263,
+    "gpu_batches_per_second": 1.0065248992692433,
+    "producer_headroom": 6.515771732407918,
     "sustainable": true,
-    "buffer_seconds": 60.36425943937202,
+    "buffer_seconds": 60.60456134198686,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 471169.052882421,
-    "upload_bytes_per_second": 5060435.640786491,
+    "generation_bytes_per_second": 470284.70092359075,
+    "upload_bytes_per_second": 5047798.77135207,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.1019104943702,
+    "estimated_full_shards_per_hour": 100.9121491506652,
     "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": [
-    {
-      "seconds": 0.8971567329990648,
-      "bytes": 7415356,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 0.9046226490008848,
       "bytes": 1040439,
@@ -562,6 +556,12 @@
       "bytes": 6269698,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9710190369987686,
+      "bytes": 6743419,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -218,3 +218,4 @@
 {"step": 61800, "loss": 0.6251155622303486, "lr": 0.0003, "grad_norm": 0.13311757053465478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9044648049, "tokens_per_second": 148496.13202418032, "data_wait_s": 7.819999154889956e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 61900, "loss": 0.6142852064222097, "lr": 0.0003, "grad_norm": 0.13330632588506205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9059131882, "tokens_per_second": 147953.8130466831, "data_wait_s": 1.0728999768616632e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 62000, "loss": 0.7418068312108517, "lr": 0.0003, "grad_norm": 0.23396590547089302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9073921292, "tokens_per_second": 150311.58204020569, "data_wait_s": 6.825001037213951e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 62100, "loss": 0.6456045061349869, "lr": 0.0003, "grad_norm": 0.14454008147655648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9088626948, "tokens_per_second": 150185.26051738468, "data_wait_s": 9.775998478289694e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3063d2790dbc028be0f51fbb79e596ad120c1e367887f6ef6c3f30d79199b25c
-size 469257
+oid sha256:c56a3e684b2304c226c007a23120b9ec11fef703dc2101143e1bfa45ca9dd26d
+size 471375