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": 76899.38131576907,
-      "required_pages_per_second_upper_estimate": 0.005320077630572572
+      "required_token_positions_per_second": 77020.5830637974,
+      "required_pages_per_second_upper_estimate": 0.005328462648727972
     },
     {
       "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": 9612.422664471134,
-      "required_pages_per_second_upper_estimate": 0.004024520639403037
+      "required_token_positions_per_second": 9627.572882974675,
+      "required_pages_per_second_upper_estimate": 0.0040308637195179305
     },
     {
       "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": 19224.84532894227,
-      "required_pages_per_second_upper_estimate": 0.00400402828875353
+      "required_token_positions_per_second": 19255.14576594935,
+      "required_pages_per_second_upper_estimate": 0.004010339070705846
     },
     {
       "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": 28837.267993413403,
-      "required_pages_per_second_upper_estimate": 0.0017287501842464406
+      "required_token_positions_per_second": 28882.718648924023,
+      "required_pages_per_second_upper_estimate": 0.0017314748816451696
     },
     {
       "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": 28837.267993413403,
-      "required_pages_per_second_upper_estimate": 0.002241970744751037
+      "required_token_positions_per_second": 28882.718648924023,
+      "required_pages_per_second_upper_estimate": 0.002245504333300682
     },
     {
       "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": 28837.267993413403,
-      "required_pages_per_second_upper_estimate": 0.3698935107735073
+      "required_token_positions_per_second": 28882.718648924023,
+      "required_pages_per_second_upper_estimate": 0.3704765029812858
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09294128937497703,
-    "producer_p95_seconds": 0.2145475670004089,
-    "gpu_update_seconds": 0.9587593390023965,
-    "producer_batches_per_second": 10.75948060033299,
-    "gpu_batches_per_second": 1.0430146120302881,
-    "producer_headroom": 10.315752508384366,
+    "producer_batch_seconds": 0.08066943803116544,
+    "producer_p95_seconds": 0.21289125200019043,
+    "gpu_update_seconds": 0.9572506084370966,
+    "producer_batches_per_second": 12.396268331678037,
+    "gpu_batches_per_second": 1.0446585159477728,
+    "producer_headroom": 11.866335402848314,
     "sustainable": true,
-    "buffer_seconds": 60.40183835715098,
+    "buffer_seconds": 60.306788331537085,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268617.8734574305,
-    "upload_bytes_per_second": 3408176.419721202,
+    "generation_bytes_per_second": 269184.3340229032,
+    "upload_bytes_per_second": 4218781.812410461,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.63914253990351,
+    "estimated_full_shards_per_hour": 57.76069179072687,
     "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": 25.62068473599993,
-      "bytes": 7109407,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.223384894999981,
       "bytes": 7577595,
@@ -562,6 +556,12 @@
       "bytes": 11719387,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0358149310013687,
+      "bytes": 6338461,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -133,3 +133,4 @@
 {"step": 698300, "loss": 0.5354017019271851, "lr": 8e-05, "grad_norm": 0.23375011422462638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102212240388, "tokens_per_second": 145206.8832564163, "data_wait_s": 1.553699985379353e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 698400, "loss": 0.5343481302261353, "lr": 8e-05, "grad_norm": 0.2049336983288719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102226991901, "tokens_per_second": 153268.80534206703, "data_wait_s": 1.8017000911640935e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 698500, "loss": 0.5305786728858948, "lr": 8e-05, "grad_norm": 0.21394717140636219, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102241679933, "tokens_per_second": 153088.39200416883, "data_wait_s": 1.3371998647926375e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 698600, "loss": 0.5597310066223145, "lr": 8e-05, "grad_norm": 0.20332029918790948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102256227774, "tokens_per_second": 151777.83470268233, "data_wait_s": 1.676499960012734e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6e085b2af1fa38819eadb12e93e4b8721ca43884323c3f322177cfe11c04cb05
-size 287661
+oid sha256:d6a08a8be761518c17d2190c1ab60998d1875ac8b0fc8873f074c9b7ffd3437d
+size 289779