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": 76937.04769164529,
-      "required_pages_per_second_upper_estimate": 0.005322683477853722
+      "required_token_positions_per_second": 77284.78742674278,
+      "required_pages_per_second_upper_estimate": 0.005346740919595116
     },
     {
       "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": 9617.130961455661,
-      "required_pages_per_second_upper_estimate": 0.004026491905030036
+      "required_token_positions_per_second": 9660.598428342848,
+      "required_pages_per_second_upper_estimate": 0.004044690825711774
     },
     {
       "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": 19234.261922911322,
-      "required_pages_per_second_upper_estimate": 0.004005989516945001
+      "required_token_positions_per_second": 19321.196856685696,
+      "required_pages_per_second_upper_estimate": 0.004024095771021827
     },
     {
       "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": 28851.392884366986,
-      "required_pages_per_second_upper_estimate": 0.0017295969498916483
+      "required_token_positions_per_second": 28981.795285028544,
+      "required_pages_per_second_upper_estimate": 0.0017374143746983718
     },
     {
       "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": 28851.392884366986,
-      "required_pages_per_second_upper_estimate": 0.0022430688928938487
+      "required_token_positions_per_second": 28981.795285028544,
+      "required_pages_per_second_upper_estimate": 0.002253207105965742
     },
     {
       "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": 28851.392884366986,
-      "required_pages_per_second_upper_estimate": 0.3700746897085336
+      "required_token_positions_per_second": 28981.795285028544,
+      "required_pages_per_second_upper_estimate": 0.37174735168903095
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1115677600468814,
-    "producer_p95_seconds": 0.2318340800011356,
-    "gpu_update_seconds": 0.9582899553865547,
-    "producer_batches_per_second": 8.963162831088429,
-    "gpu_batches_per_second": 1.043525494949616,
-    "producer_headroom": 8.589308909526155,
+    "producer_batch_seconds": 0.09190934260925587,
+    "producer_p95_seconds": 0.22100657199916895,
+    "gpu_update_seconds": 0.9539781689881179,
+    "producer_batches_per_second": 10.880286721790712,
+    "gpu_batches_per_second": 1.0482420169642848,
+    "producer_headroom": 10.379556004919635,
     "sustainable": true,
-    "buffer_seconds": 60.37226718935295,
+    "buffer_seconds": 60.10062464625143,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472157.9662748963,
-    "upload_bytes_per_second": 5223528.782039431,
+    "generation_bytes_per_second": 472085.4410031079,
+    "upload_bytes_per_second": 5242620.329022737,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.31410828766981,
+    "estimated_full_shards_per_hour": 101.29854605264595,
     "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": 0.975858148000043,
-      "bytes": 861206,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.055939247000424,
       "bytes": 7659150,
@@ -562,6 +556,12 @@
       "bytes": 12061908,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9588815639981476,
+      "bytes": 2343317,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -208,3 +208,4 @@
 {"step": 577900, "loss": 0.570175290107727, "lr": 0.0003, "grad_norm": 0.14820704823070996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84587141374, "tokens_per_second": 152481.7522259757, "data_wait_s": 1.3100001524435356e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 578000, "loss": 0.6239941716194153, "lr": 0.0003, "grad_norm": 0.1688119190484631, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84601838982, "tokens_per_second": 152260.97375723397, "data_wait_s": 8.334998710779473e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 578100, "loss": 0.5821698307991028, "lr": 0.0003, "grad_norm": 0.16252691830957244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84616504976, "tokens_per_second": 152554.61799714432, "data_wait_s": 1.166199945146218e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 578200, "loss": 0.567690908908844, "lr": 0.0003, "grad_norm": 0.1695562484367585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84631326615, "tokens_per_second": 154239.39607172104, "data_wait_s": 9.08000220078975e-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 sha256:651ff2b0b5a4b407c51f01849a52df2afbe8f31cc31e0b6b4a70d6d9b0f148da
-size 448077
+oid sha256:f81af363f94fbfeced70711275da2d4708d742704f696b09fbdfdd88872ca581
+size 450195