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": 76593.5994071842,
-      "required_pages_per_second_upper_estimate": 0.005298922928624891
+      "required_token_positions_per_second": 76766.28515100486,
+      "required_pages_per_second_upper_estimate": 0.005310869729068534
     },
     {
       "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": 9574.199925898025,
-      "required_pages_per_second_upper_estimate": 0.004008517576944022
+      "required_token_positions_per_second": 9595.785643875608,
+      "required_pages_per_second_upper_estimate": 0.004017555066300187
     },
     {
       "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": 19148.39985179605,
-      "required_pages_per_second_upper_estimate": 0.0039881067118667755
+      "required_token_positions_per_second": 19191.571287751216,
+      "required_pages_per_second_upper_estimate": 0.003997098183468909
     },
     {
       "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": 28722.599777694075,
-      "required_pages_per_second_upper_estimate": 0.0017218759998022944
+      "required_token_positions_per_second": 28787.356931626822,
+      "required_pages_per_second_upper_estimate": 0.0017257580923021138
     },
     {
       "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": 28722.599777694075,
-      "required_pages_per_second_upper_estimate": 0.0022330557953510366
+      "required_token_positions_per_second": 28787.356931626822,
+      "required_pages_per_second_upper_estimate": 0.00223809037923269
     },
     {
       "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": 28722.599777694075,
-      "required_pages_per_second_upper_estimate": 0.3684226700233973
+      "required_token_positions_per_second": 28787.356931626822,
+      "required_pages_per_second_upper_estimate": 0.3692533052632319
     }
   ],
   "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": 0.4648095479452081,
-    "producer_p95_seconds": 0.30532744100128184,
-    "gpu_update_seconds": 0.9625869598848567,
-    "producer_batches_per_second": 2.1514188002822183,
-    "gpu_batches_per_second": 1.038867179459421,
-    "producer_headroom": 2.0709276824027865,
+    "producer_batch_seconds": 0.0978031590391879,
+    "producer_p95_seconds": 0.2742331400004332,
+    "gpu_update_seconds": 0.9604216207020004,
+    "producer_batches_per_second": 10.224618609704812,
+    "gpu_batches_per_second": 1.0412093797608082,
+    "producer_headroom": 9.81994477619253,
     "sustainable": true,
-    "buffer_seconds": 60.642978472745966,
+    "buffer_seconds": 60.50656210422603,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266344.99225137284,
-    "upload_bytes_per_second": 3196925.5768788476,
+    "generation_bytes_per_second": 281953.6563057325,
+    "upload_bytes_per_second": 3198338.6932245367,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.15143514305009,
+    "estimated_full_shards_per_hour": 60.500691098012744,
     "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": 1.4273169570005848,
-      "bytes": 6151606,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4220550049976737,
       "bytes": 4624242,
@@ -562,6 +556,12 @@
       "bytes": 11753421,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.5141362069989555,
+      "bytes": 6622032,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -310,3 +310,4 @@
 {"step": 831000, "loss": 0.485998272895813, "lr": 3e-05, "grad_norm": 0.2911756141859777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121635126174, "tokens_per_second": 151405.75917899908, "data_wait_s": 1.3172000763006508e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 831100, "loss": 0.5077775716781616, "lr": 3e-05, "grad_norm": 0.28309942269289595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121649626212, "tokens_per_second": 149757.57487485392, "data_wait_s": 1.3516000763047487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 831200, "loss": 0.5240252614021301, "lr": 3e-05, "grad_norm": 0.28140193569357597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121664154969, "tokens_per_second": 150703.38592769403, "data_wait_s": 4.429999535204843e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 831300, "loss": 0.5212724804878235, "lr": 3e-05, "grad_norm": 0.2769141440323872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121678927789, "tokens_per_second": 152977.0801833194, "data_wait_s": 8.132999937515706e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ac38ca831d0051dc7dd51543357a07cfb4728021d7b224c0efa934f8df12a46
-size 665679
+oid sha256:601c1a8ab383c8431eec3d093565018f72e25e40fc6d42099c0c79b7b86ca646
+size 667797