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": 76663.6100169446,
-      "required_pages_per_second_upper_estimate": 0.005303766425055111
+      "required_token_positions_per_second": 76431.26646414833,
+      "required_pages_per_second_upper_estimate": 0.005287692358961347
     },
     {
       "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": 9582.951252118075,
-      "required_pages_per_second_upper_estimate": 0.004012181574483884
+      "required_token_positions_per_second": 9553.908308018541,
+      "required_pages_per_second_upper_estimate": 0.0040000219002750475
     },
     {
       "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": 19165.90250423615,
-      "required_pages_per_second_upper_estimate": 0.003991752052794064
+      "required_token_positions_per_second": 19107.816616037082,
+      "required_pages_per_second_upper_estimate": 0.003979654294110081
     },
     {
       "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": 28748.853756354223,
-      "required_pages_per_second_upper_estimate": 0.001723449885735466
+      "required_token_positions_per_second": 28661.724924055627,
+      "required_pages_per_second_upper_estimate": 0.0017182266452771889
     },
     {
       "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": 28748.853756354223,
-      "required_pages_per_second_upper_estimate": 0.002235096926712815
+      "required_token_positions_per_second": 28661.724924055627,
+      "required_pages_per_second_upper_estimate": 0.0022283230432408306
     },
     {
       "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": 28748.853756354223,
-      "required_pages_per_second_upper_estimate": 0.36875942787232363
+      "required_token_positions_per_second": 28661.724924055627,
+      "required_pages_per_second_upper_estimate": 0.36764183276324863
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11096670347646409,
-    "producer_p95_seconds": 0.17896080999889818,
-    "gpu_update_seconds": 0.9617079078809914,
-    "producer_batches_per_second": 9.011712240438852,
-    "gpu_batches_per_second": 1.0398167591273952,
-    "producer_headroom": 8.66663492517797,
+    "producer_batch_seconds": 0.08192778407820356,
+    "producer_p95_seconds": 0.19270915699962643,
+    "gpu_update_seconds": 0.9646314055856139,
+    "producer_batches_per_second": 12.205871442164936,
+    "gpu_batches_per_second": 1.036665398005484,
+    "producer_headroom": 11.774166925652867,
     "sustainable": true,
-    "buffer_seconds": 60.587598196502455,
+    "buffer_seconds": 60.771778551893675,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311588.6310682354,
-    "upload_bytes_per_second": 4804258.051327195,
+    "generation_bytes_per_second": 300071.0713454004,
+    "upload_bytes_per_second": 4827389.929714931,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.85966681514068,
+    "estimated_full_shards_per_hour": 64.38826661368856,
     "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": [
@@ -418,6 +418,12 @@
       "bytes": 10810645,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.2842068059999292,
+      "bytes": 7447718,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -88,3 +88,4 @@
 {"step": 728800, "loss": 0.6301681995391846, "lr": 5e-05, "grad_norm": 0.2723627800953669, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106675974294, "tokens_per_second": 131223.0555318973, "data_wait_s": 7.137001375667751e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 728900, "loss": 0.5243343710899353, "lr": 5e-05, "grad_norm": 0.209753544654001, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106690494103, "tokens_per_second": 150518.1005956999, "data_wait_s": 9.298000804847106e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 729000, "loss": 0.5009329319000244, "lr": 5e-05, "grad_norm": 0.20051109723056734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106705132571, "tokens_per_second": 151160.04991436226, "data_wait_s": 1.3786999261355959e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 729100, "loss": 0.49145078659057617, "lr": 5e-05, "grad_norm": 0.2316350361631726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106719782811, "tokens_per_second": 150992.7338657109, "data_wait_s": 1.003099896479398e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:52f836c78d75a575d77c263c859dcebd8192f633f1e9396ad2f4eb8422648605
-size 191568
+oid sha256:8d05b5bef0c2296dc2b125c80cab604bfed1ac623b73c8a65a5ce3f4655bd059
+size 193686