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": 75005.84438312017,
-      "required_pages_per_second_upper_estimate": 0.006501316490554252
+      "required_token_positions_per_second": 75619.07020211437,
+      "required_pages_per_second_upper_estimate": 0.006554469350338039
     },
     {
       "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": 9375.730547890022,
-      "required_pages_per_second_upper_estimate": 0.0037401190952170184
+      "required_token_positions_per_second": 9452.383775264296,
+      "required_pages_per_second_upper_estimate": 0.0037706972136844967
     },
     {
       "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": 18751.461095780043,
-      "required_pages_per_second_upper_estimate": 0.00433782051006482
+      "required_token_positions_per_second": 18904.76755052859,
+      "required_pages_per_second_upper_estimate": 0.004373285260269445
     },
     {
       "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": 28127.191643670067,
-      "required_pages_per_second_upper_estimate": 0.0017127983991299155
+      "required_token_positions_per_second": 28357.15132579289,
+      "required_pages_per_second_upper_estimate": 0.0017268017372659868
     },
     {
       "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": 28127.191643670067,
-      "required_pages_per_second_upper_estimate": 0.0021653090737795205
+      "required_token_positions_per_second": 28357.15132579289,
+      "required_pages_per_second_upper_estimate": 0.0021830120066784774
     },
     {
       "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": 28127.191643670067,
-      "required_pages_per_second_upper_estimate": 0.34924558455952004
+      "required_token_positions_per_second": 28357.15132579289,
+      "required_pages_per_second_upper_estimate": 0.35210091418593803
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09566370809361047,
-    "producer_p95_seconds": 0.28914204300235724,
-    "gpu_update_seconds": 0.982963402470438,
-    "producer_batches_per_second": 10.453284949204177,
-    "gpu_batches_per_second": 1.0173318736859833,
-    "producer_headroom": 10.275196540662757,
+    "producer_batch_seconds": 0.07354231682811019,
+    "producer_p95_seconds": 0.2549473239996587,
+    "gpu_update_seconds": 0.9749921521507746,
+    "producer_batches_per_second": 13.597613498324934,
+    "gpu_batches_per_second": 1.025649281170171,
+    "producer_headroom": 13.25756644884625,
     "sustainable": true,
-    "buffer_seconds": 60.943730953167154,
+    "buffer_seconds": 60.449513433348024,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 174,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 487176.0792316789,
+    "generation_bytes_per_second": 615032.5312327559,
     "upload_bytes_per_second": 4873224.191965345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.5366457244184,
+    "estimated_full_shards_per_hour": 65.98583198898795,
     "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": [

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

@@ -221,3 +221,4 @@
 {"step": 217100, "loss": 0.579597033560276, "lr": 0.0003, "grad_norm": 0.15585648008007905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31776458243, "tokens_per_second": 149510.24930505687, "data_wait_s": 1.4774999726796523e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217200, "loss": 0.5898901894688606, "lr": 0.0003, "grad_norm": 0.15459375768144146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31791071925, "tokens_per_second": 148009.81449124365, "data_wait_s": 1.0289000783814117e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217300, "loss": 0.5165498238056898, "lr": 0.0003, "grad_norm": 0.19917937040410089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31805678264, "tokens_per_second": 148768.6150525741, "data_wait_s": 1.1164000170538202e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 217400, "loss": 0.5930443629622459, "lr": 0.0003, "grad_norm": 0.14338717620554772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31820352549, "tokens_per_second": 148361.99374354762, "data_wait_s": 8.110000635497272e-06, "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 sha256:4257d0bba25a9e48692ede2dc4ff1b0709a0d6e024acdef33cacf094dc15b90c
-size 475611
+oid sha256:6ec6e3aaeaa7cf0b72456b2f06d51fca2669b171ae9f9c272b9cdb2fbfe23aed
+size 477729