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": 75381.87289876418,
-      "required_pages_per_second_upper_estimate": 0.006533909689254714
+      "required_token_positions_per_second": 75026.55163812856,
+      "required_pages_per_second_upper_estimate": 0.00650311134293632
     },
     {
       "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": 9422.734112345523,
-      "required_pages_per_second_upper_estimate": 0.0037588695198442325
+      "required_token_positions_per_second": 9378.31895476607,
+      "required_pages_per_second_upper_estimate": 0.0037411516494200055
     },
     {
       "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": 18845.468224691045,
-      "required_pages_per_second_upper_estimate": 0.004359567404869418
+      "required_token_positions_per_second": 18756.63790953214,
+      "required_pages_per_second_upper_estimate": 0.004339018074817558
     },
     {
       "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": 28268.202337036568,
-      "required_pages_per_second_upper_estimate": 0.001721385210530004
+      "required_token_positions_per_second": 28134.95686429821,
+      "required_pages_per_second_upper_estimate": 0.0017132712603252053
     },
     {
       "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": 28268.202337036568,
-      "required_pages_per_second_upper_estimate": 0.002176164467297456
+      "required_token_positions_per_second": 28134.95686429821,
+      "required_pages_per_second_upper_estimate": 0.002165906861958981
     },
     {
       "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": 28268.202337036568,
-      "required_pages_per_second_upper_estimate": 0.35099646543869983
+      "required_token_positions_per_second": 28134.95686429821,
+      "required_pages_per_second_upper_estimate": 0.3493420026111999
     }
   ],
   "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": 0.12398374739066043,
-    "producer_p95_seconds": 0.20262240800002473,
-    "gpu_update_seconds": 0.9780600715375527,
-    "producer_batches_per_second": 8.06557327912585,
-    "gpu_batches_per_second": 1.0224320868430472,
-    "producer_headroom": 7.888615178373202,
+    "producer_batch_seconds": 0.11497733950787392,
+    "producer_p95_seconds": 0.22235241700036568,
+    "gpu_update_seconds": 0.9826921055309621,
+    "producer_batches_per_second": 8.69736597037469,
+    "gpu_batches_per_second": 1.0176127338070822,
+    "producer_headroom": 8.546832878000844,
     "sustainable": true,
-    "buffer_seconds": 60.63972443532827,
+    "buffer_seconds": 60.92691054291965,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297903.1828239467,
-    "upload_bytes_per_second": 5231498.027771833,
+    "generation_bytes_per_second": 319286.4151147894,
+    "upload_bytes_per_second": 5239137.393507432,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.923088202846536,
+    "estimated_full_shards_per_hour": 68.51143207628976,
     "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": [
@@ -418,6 +418,12 @@
       "bytes": 6858339,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3612198569990142,
+      "bytes": 7511604,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -76,3 +76,4 @@
 {"step": 202600, "loss": 0.6185240875929594, "lr": 0.0003, "grad_norm": 0.15168966879030246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29652808878, "tokens_per_second": 145896.18207990756, "data_wait_s": 8.65400033944752e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 202700, "loss": 0.6121512372046709, "lr": 0.0003, "grad_norm": 0.1389628787426531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29667447656, "tokens_per_second": 148604.21248002467, "data_wait_s": 1.978900036192499e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 202800, "loss": 0.6024265885353088, "lr": 0.0003, "grad_norm": 0.14127869008033742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29682049824, "tokens_per_second": 147475.89511952884, "data_wait_s": 1.524700019217562e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 202900, "loss": 0.5920193083584309, "lr": 0.0003, "grad_norm": 0.14651542795545644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29696746421, "tokens_per_second": 149019.61192145862, "data_wait_s": 1.833300120779313e-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:fed239ec34eadbfa6f27aaecb6844838276862fff2d6315d8e3102db19cb3afc
-size 166152
+oid sha256:3ac1f4e87c92a18b1b8638e4c21f5db5e736c3648ca8afcdad9228ad26d1f95d
+size 168270