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": 75718.6148569743,
-      "required_pages_per_second_upper_estimate": 0.0052383894671975
+      "required_token_positions_per_second": 75935.28276522995,
+      "required_pages_per_second_upper_estimate": 0.005253379055829437
     },
     {
       "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": 9464.826857121787,
-      "required_pages_per_second_upper_estimate": 0.003962725357016805
+      "required_token_positions_per_second": 9491.910345653743,
+      "required_pages_per_second_upper_estimate": 0.003974064648097582
     },
     {
       "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": 18929.653714243574,
-      "required_pages_per_second_upper_estimate": 0.003942547660138172
+      "required_token_positions_per_second": 18983.820691307486,
+      "required_pages_per_second_upper_estimate": 0.003953829212981338
     },
     {
       "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": 28394.480571365362,
-      "required_pages_per_second_upper_estimate": 0.0017022057543919068
+      "required_token_positions_per_second": 28475.731036961228,
+      "required_pages_per_second_upper_estimate": 0.0017070765957421013
     },
     {
       "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": 28394.480571365362,
-      "required_pages_per_second_upper_estimate": 0.002207545970302841
+      "required_token_positions_per_second": 28475.731036961228,
+      "required_pages_per_second_upper_estimate": 0.002213862836619891
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28394.480571365362,
-      "required_pages_per_second_upper_estimate": 0.3642139091515676
+      "required_token_positions_per_second": 28475.731036961228,
+      "required_pages_per_second_upper_estimate": 0.3652561028842784
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09545354652343008,
-    "producer_p95_seconds": 0.26464155000212486,
-    "gpu_update_seconds": 0.973710363551494,
-    "producer_batches_per_second": 10.476300110594,
-    "gpu_batches_per_second": 1.0269994419619994,
-    "producer_headroom": 10.20088198936104,
+    "producer_batch_seconds": 0.11607444742969619,
+    "producer_p95_seconds": 0.27145608499995433,
+    "gpu_update_seconds": 0.9709320531266838,
+    "producer_batches_per_second": 8.615160546903992,
+    "gpu_batches_per_second": 1.029938188547498,
+    "producer_headroom": 8.364735517821497,
     "sustainable": true,
-    "buffer_seconds": 60.37004254019263,
+    "buffer_seconds": 60.1977872938544,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 243643.97194668982,
-    "upload_bytes_per_second": 5295909.316422317,
+    "generation_bytes_per_second": 279903.7494057367,
+    "upload_bytes_per_second": 5329241.871989678,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.28032463837167,
+    "estimated_full_shards_per_hour": 60.060828796664005,
     "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": 1.6179820170000312,
-      "bytes": 6082981,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.991337299998122,
       "bytes": 1324376,
@@ -562,6 +556,12 @@
       "bytes": 947216,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.4317332190003071,
+      "bytes": 7796978,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -292,3 +292,4 @@
 {"step": 749200, "loss": 0.5976563692092896, "lr": 5e-05, "grad_norm": 0.24472929251236103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109661998612, "tokens_per_second": 152651.8169018697, "data_wait_s": 1.2212000001454726e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 749300, "loss": 0.5639749765396118, "lr": 5e-05, "grad_norm": 0.2388463105549386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109676619485, "tokens_per_second": 150578.40930742794, "data_wait_s": 1.6096000763354823e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 749400, "loss": 0.497047483921051, "lr": 5e-05, "grad_norm": 0.2345468090923336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109691158136, "tokens_per_second": 149560.74796764064, "data_wait_s": 8.914998034015298e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 749500, "loss": 0.5131856799125671, "lr": 5e-05, "grad_norm": 0.2493631220042696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109705650624, "tokens_per_second": 146846.81138012517, "data_wait_s": 1.6113001038320363e-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 sha256:40f8fda8e5709cbe31053126f12e5e55c463b25665dee9751b14ac85a1143375
-size 626772
+oid sha256:63db1cf400c38bfb1949ade283703227f0dffed7651889ec1ed78b3ee8554195
+size 628890