Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74977.14641364208,
-      "required_pages_per_second_upper_estimate": 0.00649882902329423
+      "required_token_positions_per_second": 75234.18819853998,
+      "required_pages_per_second_upper_estimate": 0.006521108753742732
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9372.14330170526,
-      "required_pages_per_second_upper_estimate": 0.003738688089079807
+      "required_token_positions_per_second": 9404.273524817498,
+      "required_pages_per_second_upper_estimate": 0.0037515053154689236
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18744.28660341052,
-      "required_pages_per_second_upper_estimate": 0.004336160817521884
+      "required_token_positions_per_second": 18808.547049634995,
+      "required_pages_per_second_upper_estimate": 0.00435102634081069
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28116.429905115783,
-      "required_pages_per_second_upper_estimate": 0.0017121430657144382
+      "required_token_positions_per_second": 28212.820574452493,
+      "required_pages_per_second_upper_estimate": 0.001718012751754286
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28116.429905115783,
-      "required_pages_per_second_upper_estimate": 0.0021644806053552124
+      "required_token_positions_per_second": 28212.820574452493,
+      "required_pages_per_second_upper_estimate": 0.002171901025907203
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28116.429905115783,
-      "required_pages_per_second_upper_estimate": 0.34911195978389786
+      "required_token_positions_per_second": 28212.820574452493,
+      "required_pages_per_second_upper_estimate": 0.3503088092982417
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.2129938897969481,
-    "producer_p95_seconds": 0.21051874400109227,
-    "gpu_update_seconds": 0.9833396378310978,
-    "producer_batches_per_second": 4.6949703625457175,
-    "gpu_batches_per_second": 1.016942632563505,
-    "producer_headroom": 4.616750455933444,
+    "producer_batch_seconds": 0.08789954166407199,
+    "producer_p95_seconds": 0.2354176049993839,
+    "gpu_update_seconds": 0.9799800033122547,
+    "producer_batches_per_second": 11.376623598581737,
+    "gpu_batches_per_second": 1.020428984897732,
+    "producer_headroom": 11.148863631820404,
     "sustainable": true,
-    "buffer_seconds": 60.967057545528064,
+    "buffer_seconds": 60.75876020535979,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 427551.7937554527,
-    "upload_bytes_per_second": 5366563.81582847,
+    "generation_bytes_per_second": 431558.9438629053,
+    "upload_bytes_per_second": 5316446.155773242,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.7426620435494,
+    "estimated_full_shards_per_hour": 92.60250317492837,
     "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": 1.2444776140000613,
-      "bytes": 6243354,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.5528372400003718,
       "bytes": 12775308,
@@ -562,6 +556,12 @@
       "bytes": 11061746,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.9865021240002534,
+      "bytes": 862370,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -130,3 +130,4 @@
 {"step": 208000, "loss": 0.607620956376195, "lr": 0.0003, "grad_norm": 0.14732116012454316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30443630127, "tokens_per_second": 148512.86461284247, "data_wait_s": 1.696700019238051e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 208100, "loss": 0.6044241823256016, "lr": 0.0003, "grad_norm": 0.1422607636402442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30458311258, "tokens_per_second": 148065.74900817507, "data_wait_s": 2.005399983318057e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 208200, "loss": 0.5959686599671841, "lr": 0.0003, "grad_norm": 0.17725419329156736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30472798873, "tokens_per_second": 146882.4358199653, "data_wait_s": 2.543499977036845e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 208300, "loss": 0.5849227532744408, "lr": 0.0003, "grad_norm": 0.1436915649278191, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30487336663, "tokens_per_second": 146661.76946535383, "data_wait_s": 1.5249999705702066e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3466c563cb5b14a9118ebf268aa09b2bcb3e08fd88e0656c9c4af5473e57948
-size 281307
+oid sha256:c53381207c524db1f60240080edeaf8409ec1cb63742bd11f9d15bcc5e9fee4f
+size 283425