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": 76293.90301586306,
-      "required_pages_per_second_upper_estimate": 0.006612962148285647
+      "required_token_positions_per_second": 76228.00974719698,
+      "required_pages_per_second_upper_estimate": 0.006607250686762633
     },
     {
       "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": 9536.737876982883,
-      "required_pages_per_second_upper_estimate": 0.003804347326066253
+      "required_token_positions_per_second": 9528.501218399622,
+      "required_pages_per_second_upper_estimate": 0.00380106159980837
     },
     {
       "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": 19073.475753965766,
-      "required_pages_per_second_upper_estimate": 0.004412312933971664
+      "required_token_positions_per_second": 19057.002436799245,
+      "required_pages_per_second_upper_estimate": 0.004408502121965669
     },
     {
       "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": 28610.21363094865,
-      "required_pages_per_second_upper_estimate": 0.0017422119039346687
+      "required_token_positions_per_second": 28585.503655198867,
+      "required_pages_per_second_upper_estimate": 0.0017407071960547304
     },
     {
       "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": 28610.21363094865,
-      "required_pages_per_second_upper_estimate": 0.002202493443450664
+      "required_token_positions_per_second": 28585.503655198867,
+      "required_pages_per_second_upper_estimate": 0.002200591201115855
     },
     {
       "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": 28610.21363094865,
-      "required_pages_per_second_upper_estimate": 0.35524310107091955
+      "required_token_positions_per_second": 28585.503655198867,
+      "required_pages_per_second_upper_estimate": 0.35493628587107623
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09977362698458592,
-    "producer_p95_seconds": 0.3880841270001838,
-    "gpu_update_seconds": 0.9663681773453173,
-    "producer_batches_per_second": 10.022688662550982,
-    "gpu_batches_per_second": 1.0348022869990108,
-    "producer_headroom": 9.685607374928969,
+    "producer_batch_seconds": 0.11422511410142988,
+    "producer_p95_seconds": 0.3618045629991684,
+    "gpu_update_seconds": 0.9672035285259575,
+    "producer_batches_per_second": 8.754642163123757,
+    "gpu_batches_per_second": 1.033908552343709,
+    "producer_headroom": 8.467520791155419,
     "sustainable": true,
-    "buffer_seconds": 60.88119517275499,
+    "buffer_seconds": 60.93382229713532,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354749.98800143506,
+    "generation_bytes_per_second": 354783.7727048669,
     "upload_bytes_per_second": 5691532.768220302,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.12108926803865,
+    "estimated_full_shards_per_hour": 76.12833867892746,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -345,3 +345,4 @@
 {"step": 314700, "loss": 0.6050030756741762, "lr": 0.0003, "grad_norm": 0.16552769369539239, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46063448200, "tokens_per_second": 150341.50362639906, "data_wait_s": 1.4864002878312021e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 314800, "loss": 0.662660850211978, "lr": 0.0003, "grad_norm": 0.15615778817110823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46078226255, "tokens_per_second": 150534.09911022458, "data_wait_s": 1.2757998774759471e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 314900, "loss": 0.5819839462637901, "lr": 0.0003, "grad_norm": 0.13615964844314887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46092755066, "tokens_per_second": 149192.2055798279, "data_wait_s": 9.184994269162416e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 315000, "loss": 0.5823196284472942, "lr": 0.0003, "grad_norm": 0.1495047343906294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46107274235, "tokens_per_second": 148961.74686176443, "data_wait_s": 1.3094999303575605e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f4529d8f3bf83952b20a165654105a490eab239a1494c2affeeb2a49d013084
-size 739809
+oid sha256:accd60dbe55dc48ac2a74edafed47e067c249be715b849812532d15e08ca2dbd
+size 741927