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": 76724.93157536862,
-      "required_pages_per_second_upper_estimate": 0.005308008792752503
+      "required_token_positions_per_second": 76570.19478512557,
+      "required_pages_per_second_upper_estimate": 0.005297303742564669
     },
     {
       "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": 9590.616446921078,
-      "required_pages_per_second_upper_estimate": 0.004015390831480432
+      "required_token_positions_per_second": 9571.274348140696,
+      "required_pages_per_second_upper_estimate": 0.004007292698630038
     },
     {
       "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": 19181.232893842156,
-      "required_pages_per_second_upper_estimate": 0.003994944968659433
+      "required_token_positions_per_second": 19142.548696281392,
+      "required_pages_per_second_upper_estimate": 0.003986888070478419
     },
     {
       "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": 28771.849340763234,
-      "required_pages_per_second_upper_estimate": 0.001724828435908562
+      "required_token_positions_per_second": 28713.823044422086,
+      "required_pages_per_second_upper_estimate": 0.0017213498480439344
     },
     {
       "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": 28771.849340763234,
-      "required_pages_per_second_upper_estimate": 0.0022368847322537296
+      "required_token_positions_per_second": 28713.823044422086,
+      "required_pages_per_second_upper_estimate": 0.002232373442943909
     },
     {
       "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": 28771.849340763234,
-      "required_pages_per_second_upper_estimate": 0.36905439053838757
+      "required_token_positions_per_second": 28713.823044422086,
+      "required_pages_per_second_upper_estimate": 0.36831009151270616
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.13739597826582894,
-    "producer_p95_seconds": 0.30908698799976264,
-    "gpu_update_seconds": 0.9609392734039174,
-    "producer_batches_per_second": 7.278233414265117,
-    "gpu_batches_per_second": 1.0406484859940406,
-    "producer_headroom": 6.993940328768034,
+    "producer_batch_seconds": 0.10556576354682079,
+    "producer_p95_seconds": 0.31098492999808514,
+    "gpu_update_seconds": 0.9628811864315946,
+    "producer_batches_per_second": 9.472768124833175,
+    "gpu_batches_per_second": 1.0385497339562386,
+    "producer_headroom": 9.12115021083076,
     "sustainable": true,
-    "buffer_seconds": 60.539174224446796,
+    "buffer_seconds": 60.66151474519046,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350176.6534592285,
-    "upload_bytes_per_second": 6181980.00635588,
+    "generation_bytes_per_second": 407733.9331541559,
+    "upload_bytes_per_second": 6128464.793724774,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.13975813706057,
+    "estimated_full_shards_per_hour": 87.49021049469478,
     "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": 0.9754138270000112,
-      "bytes": 5934822,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.7901811400006409,
       "bytes": 1002646,
@@ -562,6 +556,12 @@
       "bytes": 6276750,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0008894900020096,
+      "bytes": 2319506,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -286,3 +286,4 @@
 {"step": 513600, "loss": 0.6071254014968872, "lr": 0.0003, "grad_norm": 0.1688218516042935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75173572909, "tokens_per_second": 151571.5938336054, "data_wait_s": 1.671099744271487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 513700, "loss": 0.5523330569267273, "lr": 0.0003, "grad_norm": 0.17923486800237254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75188151363, "tokens_per_second": 151950.55079426762, "data_wait_s": 2.088300243485719e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 513800, "loss": 0.6258680820465088, "lr": 0.0003, "grad_norm": 0.16624693792826142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75202696003, "tokens_per_second": 151591.762574677, "data_wait_s": 1.395299841533415e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 513900, "loss": 0.5715893507003784, "lr": 0.0003, "grad_norm": 0.18317833144933246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75217419312, "tokens_per_second": 153774.40437191527, "data_wait_s": 9.772997145773843e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:11fd1f325256562ca9a804848cf4ff18dd21cfa104ef35ddf8b340f1dc9f45d3
-size 614064
+oid sha256:6bf55fdacc06ac17ba4c63091ec6471f1c9768fa16bd88c0621e72f8df5a2830
+size 616182