Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76632.7813192813,
-      "required_pages_per_second_upper_estimate": 0.00530163362421833
+      "required_token_positions_per_second": 76191.18383289281,
+      "required_pages_per_second_upper_estimate": 0.005271082885462637
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9579.097664910163,
-      "required_pages_per_second_upper_estimate": 0.004010568157991983
+      "required_token_positions_per_second": 9523.897979111602,
+      "required_pages_per_second_upper_estimate": 0.003987457202248642
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19158.195329820326,
-      "required_pages_per_second_upper_estimate": 0.003990146851615105
+      "required_token_positions_per_second": 19047.795958223203,
+      "required_pages_per_second_upper_estimate": 0.003967153573938638
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28737.292994730487,
-      "required_pages_per_second_upper_estimate": 0.001722756835728384
+      "required_token_positions_per_second": 28571.693937334807,
+      "required_pages_per_second_upper_estimate": 0.0017128294250926329
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28737.292994730487,
-      "required_pages_per_second_upper_estimate": 0.002234198128346986
+      "required_token_positions_per_second": 28571.693937334807,
+      "required_pages_per_second_upper_estimate": 0.0022213235300277464
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28737.292994730487,
-      "required_pages_per_second_upper_estimate": 0.3686111388351931
+      "required_token_positions_per_second": 28571.693937334807,
+      "required_pages_per_second_upper_estimate": 0.366487011933336
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.04736038308592683,
-    "producer_p95_seconds": 0.14617853599975206,
-    "gpu_update_seconds": 0.9620947945608437,
-    "producer_batches_per_second": 21.114694072167477,
-    "gpu_batches_per_second": 1.0393986181543144,
-    "producer_headroom": 20.314337255577033,
+    "producer_batch_seconds": 0.0898157678984326,
+    "producer_p95_seconds": 0.15151765500013425,
+    "gpu_update_seconds": 0.9676710124586695,
+    "producer_batches_per_second": 11.133902469451039,
+    "gpu_batches_per_second": 1.0334090689140194,
+    "producer_headroom": 10.773954675229767,
     "sustainable": true,
-    "buffer_seconds": 60.611972057333155,
+    "buffer_seconds": 60.96327378489618,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287029.09276674007,
-    "upload_bytes_per_second": 5137036.082966505,
+    "generation_bytes_per_second": 288518.3979374297,
+    "upload_bytes_per_second": 5014246.029016892,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.58976161243106,
+    "estimated_full_shards_per_hour": 61.909331832811056,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 10464030,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.128213264000351,
+      "bytes": 2552181,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 606700, "loss": 0.5418317914009094, "lr": 0.0001, "grad_norm": 0.16878725848738357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88802910141, "tokens_per_second": 151640.92633213598, "data_wait_s": 1.0528000075282762e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 606800, "loss": 0.5774630308151245, "lr": 0.0001, "grad_norm": 0.1533205028361282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88817671208, "tokens_per_second": 153215.73447032133, "data_wait_s": 1.502800023445161e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 606900, "loss": 0.5948517918586731, "lr": 0.0001, "grad_norm": 0.1392813678864415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88832465197, "tokens_per_second": 153589.1441208933, "data_wait_s": 1.406699993822258e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 607000, "loss": 0.5789181590080261, "lr": 0.0001, "grad_norm": 0.13183917411198898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88847190428, "tokens_per_second": 152226.69748541186, "data_wait_s": 9.513999884802615e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ae25160df7060eb38183453612bfafcd2cf0a2af29466686f63b128ee0b4a68
-size 53898
+oid sha256:0f95686b0b9d9549ecc15ed7b9f17a1922ce6e87be49a87ea2753eb28fd27bf6
+size 56016