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": 75488.56392437725,
-      "required_pages_per_second_upper_estimate": 0.0065431573969489225
+      "required_token_positions_per_second": 75645.3578037919,
+      "required_pages_per_second_upper_estimate": 0.006556747893026131
     },
     {
       "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": 9436.070490547156,
-      "required_pages_per_second_upper_estimate": 0.003764189600505487
+      "required_token_positions_per_second": 9455.669725473988,
+      "required_pages_per_second_upper_estimate": 0.0037720080283524765
     },
     {
       "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": 18872.14098109431,
-      "required_pages_per_second_upper_estimate": 0.0043657376776388345
+      "required_token_positions_per_second": 18911.339450947977,
+      "required_pages_per_second_upper_estimate": 0.004374805553769972
     },
     {
       "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": 28308.211471641465,
-      "required_pages_per_second_upper_estimate": 0.001723821556915736
+      "required_token_positions_per_second": 28367.009176421965,
+      "required_pages_per_second_upper_estimate": 0.0017274020286491526
     },
     {
       "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": 28308.211471641465,
-      "required_pages_per_second_upper_estimate": 0.002179244481231714
+      "required_token_positions_per_second": 28367.009176421965,
+      "required_pages_per_second_upper_estimate": 0.0021837708913081807
     },
     {
       "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": 28308.211471641465,
-      "required_pages_per_second_upper_estimate": 0.3514932449885328
+      "required_token_positions_per_second": 28367.009176421965,
+      "required_pages_per_second_upper_estimate": 0.3522233156986474
     }
   ],
   "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": 0.08863170175000334,
-    "producer_p95_seconds": 0.2126851870016253,
-    "gpu_update_seconds": 0.9766777398740697,
-    "producer_batches_per_second": 11.282644699981317,
-    "gpu_batches_per_second": 1.023879176491662,
-    "producer_headroom": 11.019507925379903,
+    "producer_batch_seconds": 0.06372443303902742,
+    "producer_p95_seconds": 0.2250764780001191,
+    "gpu_update_seconds": 0.9746533315532048,
+    "producer_batches_per_second": 15.6925680199863,
+    "gpu_batches_per_second": 1.0260058295870214,
+    "producer_headroom": 15.294813701304925,
     "sustainable": true,
-    "buffer_seconds": 60.55401987219232,
+    "buffer_seconds": 60.42850655629869,
     "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": 282894.61719676596,
-    "upload_bytes_per_second": 6006243.388537014,
+    "generation_bytes_per_second": 282493.3255318214,
+    "upload_bytes_per_second": 5952913.5623308495,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.70259940078005,
+    "estimated_full_shards_per_hour": 60.61649155107481,
     "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": 1.2109743800001525,
-      "bytes": 6387518,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.4837992330003544,
       "bytes": 17296274,
@@ -562,6 +556,12 @@
       "bytes": 12068565,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 0.9089816479990986,
+      "bytes": 779851,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 129800, "loss": 0.5880979876965284, "lr": 0.0003, "grad_norm": 0.1268623129422515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18995416582, "tokens_per_second": 142832.72843818384, "data_wait_s": 1.3171000318834558e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 129900, "loss": 0.603065631352365, "lr": 0.0003, "grad_norm": 0.1387297366005279, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19010098040, "tokens_per_second": 134676.9472382548, "data_wait_s": 7.50599974708166e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 130000, "loss": 0.6065531261265278, "lr": 0.0003, "grad_norm": 0.15030424920394553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19024827314, "tokens_per_second": 150786.94044244278, "data_wait_s": 1.2477999916882254e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 130100, "loss": 0.6317621618509293, "lr": 0.0003, "grad_norm": 0.15858009652001684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19039552814, "tokens_per_second": 141480.0370326361, "data_wait_s": 1.3619999663205817e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cf51071ae967ceba0f567997cb46e04ab3bb0c1cdfb3ba18471373b1b554656c
-size 320214
+oid sha256:24ad53b14078f559767776d418d363335d39155566f3c589404aee6bbedbb182
+size 322332