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": 75973.09294624932,
-      "required_pages_per_second_upper_estimate": 0.006585155144537215
+      "required_token_positions_per_second": 75784.17167252276,
+      "required_pages_per_second_upper_estimate": 0.006568779927347188
     },
     {
       "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": 9496.636618281165,
-      "required_pages_per_second_upper_estimate": 0.00378835033440289
+      "required_token_positions_per_second": 9473.021459065345,
+      "required_pages_per_second_upper_estimate": 0.003778929894313605
     },
     {
       "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": 18993.27323656233,
-      "required_pages_per_second_upper_estimate": 0.004393759493086477
+      "required_token_positions_per_second": 18946.04291813069,
+      "required_pages_per_second_upper_estimate": 0.004382833590142531
     },
     {
       "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": 28489.9098548435,
-      "required_pages_per_second_upper_estimate": 0.0017348860351549978
+      "required_token_positions_per_second": 28419.064377196035,
+      "required_pages_per_second_upper_estimate": 0.001730571917263762
     },
     {
       "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": 28489.9098548435,
-      "required_pages_per_second_upper_estimate": 0.002193232125743923
+      "required_token_positions_per_second": 28419.064377196035,
+      "required_pages_per_second_upper_estimate": 0.0021877782447617873
     },
     {
       "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": 28489.9098548435,
-      "required_pages_per_second_upper_estimate": 0.35374933080253174
+      "required_token_positions_per_second": 28419.064377196035,
+      "required_pages_per_second_upper_estimate": 0.35286966707471146
     }
   ],
   "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": 0.07423637789838722,
-    "producer_p95_seconds": 0.284192440998595,
-    "gpu_update_seconds": 0.9704488410411601,
-    "producer_batches_per_second": 13.470484798824284,
-    "gpu_batches_per_second": 1.0304510219489114,
-    "producer_headroom": 13.07241636128159,
+    "producer_batch_seconds": 0.09151747209378414,
+    "producer_p95_seconds": 0.29312572900016676,
+    "gpu_update_seconds": 0.9728680590267866,
+    "producer_batches_per_second": 10.926875241650386,
+    "gpu_batches_per_second": 1.0278886131798335,
+    "producer_headroom": 10.63040790757226,
     "sustainable": true,
-    "buffer_seconds": 60.167828144551926,
+    "buffer_seconds": 60.31781965966077,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 176,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 526178.4613345308,
-    "upload_bytes_per_second": 2659112.9632319887,
+    "generation_bytes_per_second": 731467.5103090211,
+    "upload_bytes_per_second": 2658643.508212608,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.90564899470276,
+    "estimated_full_shards_per_hour": 78.47795001007545,
     "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": 0.804579094001383,
-      "bytes": 1179275,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.2755006959996535,
       "bytes": 7429804,
@@ -562,6 +556,12 @@
       "bytes": 7389964,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.8301882180021494,
+      "bytes": 1174116,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -189,3 +189,4 @@
 {"step": 299100, "loss": 0.5828564874827862, "lr": 0.0003, "grad_norm": 0.15508277199096457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43780718355, "tokens_per_second": 149694.1737190229, "data_wait_s": 1.4501998521154746e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 299200, "loss": 0.6108491159975529, "lr": 0.0003, "grad_norm": 0.17647562904111608, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43795281901, "tokens_per_second": 148951.94612153881, "data_wait_s": 9.518997103441507e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 299300, "loss": 0.5905456878244877, "lr": 0.0003, "grad_norm": 0.15503655680424624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43809779987, "tokens_per_second": 148395.4664581959, "data_wait_s": 9.308998414780945e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 299400, "loss": 0.6056878976523876, "lr": 0.0003, "grad_norm": 0.14099047461920827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43824464906, "tokens_per_second": 150362.90423187314, "data_wait_s": 1.2349002645350993e-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 sha256:8467f4790334c46abc0e9bcb2f09e804c1104506820607c6622db42fe48db8c6
-size 407052
+oid sha256:2579210e25cefe2593224d114e664f2db7866209f1c7cfe93ba37cd3f713c6ad
+size 409170