Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77196.23281663316,
-      "required_pages_per_second_upper_estimate": 0.005340614506192728
+      "required_token_positions_per_second": 77499.76419256114,
+      "required_pages_per_second_upper_estimate": 0.00536161351106925
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9649.529102079145,
-      "required_pages_per_second_upper_estimate": 0.00404005632995898
+      "required_token_positions_per_second": 9687.470524070142,
+      "required_pages_per_second_upper_estimate": 0.004055941611039623
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19299.05820415829,
-      "required_pages_per_second_upper_estimate": 0.004019484873535897
+      "required_token_positions_per_second": 19374.941048140285,
+      "required_pages_per_second_upper_estimate": 0.004035289268772178
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28948.587306237434,
-      "required_pages_per_second_upper_estimate": 0.0017354236070754926
+      "required_token_positions_per_second": 29062.41157221043,
+      "required_pages_per_second_upper_estimate": 0.0017422471980209828
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28948.587306237434,
-      "required_pages_per_second_upper_estimate": 0.002250625331681195
+      "required_token_positions_per_second": 29062.41157221043,
+      "required_pages_per_second_upper_estimate": 0.002259474667700561
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28948.587306237434,
-      "required_pages_per_second_upper_estimate": 0.37132139539304826
+      "required_token_positions_per_second": 29062.41157221043,
+      "required_pages_per_second_upper_estimate": 0.37278141086197497
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.23572977017192898,
-    "producer_p95_seconds": 0.23726008599987836,
-    "gpu_update_seconds": 0.9550725120891408,
-    "producer_batches_per_second": 4.242145568931121,
-    "gpu_batches_per_second": 1.0470409181943516,
-    "producer_headroom": 4.051556625166863,
+    "producer_batch_seconds": 0.06790260375785806,
+    "producer_p95_seconds": 0.21508375700068427,
+    "gpu_update_seconds": 0.9513319268534861,
+    "producer_batches_per_second": 14.726975766143202,
+    "gpu_batches_per_second": 1.051157825962472,
+    "producer_headroom": 14.010242232329608,
     "sustainable": true,
-    "buffer_seconds": 60.16956826161587,
+    "buffer_seconds": 60.88524331862311,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 487306.6980747318,
-    "upload_bytes_per_second": 5209536.749183701,
+    "generation_bytes_per_second": 487795.98705164494,
+    "upload_bytes_per_second": 5218886.950477249,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.56467348748652,
+    "estimated_full_shards_per_hour": 104.66966351186763,
     "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": 2.1566952029988897,
-      "bytes": 6925851,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.9256938060007087,
       "bytes": 6110229,
@@ -562,6 +556,12 @@
       "bytes": 11138063,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 0.9064097570008016,
+      "bytes": 1171302,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 578300, "loss": 0.564687967300415, "lr": 0.0003, "grad_norm": 0.15118391940111528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84646007738, "tokens_per_second": 153030.38577464016, "data_wait_s": 1.004899968393147e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 578400, "loss": 0.5859547853469849, "lr": 0.0003, "grad_norm": 0.16622298513679798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84660777057, "tokens_per_second": 154022.36304698404, "data_wait_s": 7.766000635456294e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 578500, "loss": 0.5562648177146912, "lr": 0.0003, "grad_norm": 0.14849490008911154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84675307075, "tokens_per_second": 151294.98359596176, "data_wait_s": 1.0092000593431294e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 578600, "loss": 0.5609111189842224, "lr": 0.0003, "grad_norm": 0.17871318168536346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84689857947, "tokens_per_second": 151580.5485364575, "data_wait_s": 9.054001566255465e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c141826f54cbf769cafeb7945de24feb1946af0ef09b34fdea3418c56c68309
-size 456549
+oid sha256:048a2c7fd032a2d07f5198bb983a283bb82ce03ecfad3cf53248d3897f53d506
+size 458667