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": 75960.28225484007,
-      "required_pages_per_second_upper_estimate": 0.005255108578527473
+      "required_token_positions_per_second": 77134.45089773306,
+      "required_pages_per_second_upper_estimate": 0.005336340289689417
     },
     {
       "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": 9495.035281855009,
-      "required_pages_per_second_upper_estimate": 0.003975372993628963
+      "required_token_positions_per_second": 9641.806362216632,
+      "required_pages_per_second_upper_estimate": 0.004036822980047692
     },
     {
       "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": 18990.070563710018,
-      "required_pages_per_second_upper_estimate": 0.003955130896582567
+      "required_token_positions_per_second": 19283.612724433264,
+      "required_pages_per_second_upper_estimate": 0.004016267987433865
     },
     {
       "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": 28485.105845565027,
-      "required_pages_per_second_upper_estimate": 0.0017076386012033953
+      "required_token_positions_per_second": 28925.419086649898,
+      "required_pages_per_second_upper_estimate": 0.0017340347076870426
     },
     {
       "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": 28485.105845565027,
-      "required_pages_per_second_upper_estimate": 0.002214591686753411
+      "required_token_positions_per_second": 28925.419086649898,
+      "required_pages_per_second_upper_estimate": 0.0022488241045144925
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28485.105845565027,
-      "required_pages_per_second_upper_estimate": 0.3653763528631627
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28925.419086649898,
+      "required_pages_per_second_upper_estimate": 0.37102421834827537
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10261963308602162,
-    "producer_p95_seconds": 0.35040617499907967,
-    "gpu_update_seconds": 0.9706125071079783,
-    "producer_batches_per_second": 9.744723986313058,
-    "gpu_batches_per_second": 1.0302772658262813,
-    "producer_headroom": 9.45835097943057,
+    "producer_batch_seconds": 0.12034691545312626,
+    "producer_p95_seconds": 0.36695080999925267,
+    "gpu_update_seconds": 0.955837490795787,
+    "producer_batches_per_second": 8.30931142883748,
+    "gpu_batches_per_second": 1.046202947289131,
+    "producer_headroom": 7.942351386380772,
     "sustainable": true,
-    "buffer_seconds": 60.17797544069465,
+    "buffer_seconds": 60.21776192013458,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360488.13000534975,
-    "upload_bytes_per_second": 5040918.656321026,
+    "generation_bytes_per_second": 406275.0365491392,
+    "upload_bytes_per_second": 5169866.075630414,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.35236096496935,
+    "estimated_full_shards_per_hour": 87.17716524463302,
     "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": 4.141720985000575,
-      "bytes": 6233393,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9377798560017254,
       "bytes": 884750,
@@ -562,6 +556,12 @@
       "bytes": 5813649,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0730638719978742,
+      "bytes": 1187099,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -391,3 +391,4 @@
 {"step": 479300, "loss": 0.5835630893707275, "lr": 0.0003, "grad_norm": 0.15958728987571263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70154115165, "tokens_per_second": 152735.8935683535, "data_wait_s": 1.2071999663021415e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 479400, "loss": 0.5846356749534607, "lr": 0.0003, "grad_norm": 0.16122760689946794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70168834542, "tokens_per_second": 152203.28829834406, "data_wait_s": 1.4932993508409709e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 479500, "loss": 0.6068574786186218, "lr": 0.0003, "grad_norm": 0.15918015046442474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70183623826, "tokens_per_second": 153771.75652279612, "data_wait_s": 1.11860063043423e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 479600, "loss": 0.5098310708999634, "lr": 0.0003, "grad_norm": 0.16132954571477512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70198198125, "tokens_per_second": 151790.60705393672, "data_wait_s": 1.2163000064902008e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8254de1a2db3e5c50ef1bdd9d3d8b17ce9e1cc4431f59110cc5e3383dedb4cd0
-size 838020
+oid sha256:ffbe4c5f078004d64eb5256848cc6c2a8b5731835047857ea74388664db6f55a
+size 840138