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": 75235.14208509204,
-      "required_pages_per_second_upper_estimate": 0.00652119143421146
+      "required_token_positions_per_second": 75747.4214829575,
+      "required_pages_per_second_upper_estimate": 0.0065655945140581755
     },
     {
       "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": 9404.392760636505,
-      "required_pages_per_second_upper_estimate": 0.00375155288041986
+      "required_token_positions_per_second": 9468.427685369688,
+      "required_pages_per_second_upper_estimate": 0.00377709736930337
     },
     {
       "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": 18808.78552127301,
-      "required_pages_per_second_upper_estimate": 0.004351081507027186
+      "required_token_positions_per_second": 18936.855370739377,
+      "required_pages_per_second_upper_estimate": 0.004380708212748862
     },
     {
       "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": 28213.178281909517,
-      "required_pages_per_second_upper_estimate": 0.0017180345342616729
+      "required_token_positions_per_second": 28405.283056109067,
+      "required_pages_per_second_upper_estimate": 0.0017297327071145695
     },
     {
       "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": 28213.178281909517,
-      "required_pages_per_second_upper_estimate": 0.0021719285632174424
+      "required_token_positions_per_second": 28405.283056109067,
+      "required_pages_per_second_upper_estimate": 0.002186717320515374
     },
     {
       "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": 28213.178281909517,
-      "required_pages_per_second_upper_estimate": 0.3503132508276881
+      "required_token_positions_per_second": 28405.283056109067,
+      "required_pages_per_second_upper_estimate": 0.35269854918992594
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.1936005467264863,
-    "producer_p95_seconds": 0.3984401549969334,
-    "gpu_update_seconds": 0.9799675784038868,
-    "producer_batches_per_second": 5.165274669460378,
-    "gpu_batches_per_second": 1.0204419228121209,
-    "producer_headroom": 5.061801709622023,
+    "producer_batch_seconds": 0.1448010794923107,
+    "producer_p95_seconds": 0.3972985120053636,
+    "gpu_update_seconds": 0.9733400630223187,
+    "producer_batches_per_second": 6.906025863247121,
+    "gpu_batches_per_second": 1.0273901568326484,
+    "producer_headroom": 6.721911648966716,
     "sustainable": true,
-    "buffer_seconds": 60.75798986104098,
+    "buffer_seconds": 60.347083907383755,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346426.5401600312,
-    "upload_bytes_per_second": 4035684.3537830836,
+    "generation_bytes_per_second": 346526.5810567709,
+    "upload_bytes_per_second": 3987585.4294902612,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.33507112122251,
+    "estimated_full_shards_per_hour": 74.35653756882998,
     "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": 1.2849263329990208,
-      "bytes": 11334677,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.0467089929989015,
       "bytes": 955823,
@@ -562,6 +556,12 @@
       "bytes": 7314912,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1214634670031955,
+      "bytes": 5616787,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -344,3 +344,4 @@
 {"step": 349400, "loss": 0.6577212065458298, "lr": 0.0003, "grad_norm": 0.21542801983961626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51143129490, "tokens_per_second": 148095.95266899132, "data_wait_s": 1.410899858456105e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 349500, "loss": 0.6470698090270162, "lr": 0.0003, "grad_norm": 0.16589583725746204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51157762813, "tokens_per_second": 148381.10803881442, "data_wait_s": 1.894300658022985e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 349600, "loss": 0.5855649448931217, "lr": 0.0003, "grad_norm": 0.15140274884820604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51172289562, "tokens_per_second": 148063.85542529196, "data_wait_s": 1.754800177877769e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 349700, "loss": 0.5897213071584702, "lr": 0.0003, "grad_norm": 0.149293986108801, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51187014457, "tokens_per_second": 149374.86428809058, "data_wait_s": 1.1515003279782832e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea35cf336ff8384cf6c28b1d09cb84311452fd665f77c99682ae151b7e74f865
-size 737691
+oid sha256:ae7e22aa9ce0aa7b3497587c03f515d9ce1d1fee401a70e8f62830b46d983c6a
+size 739809