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": 74950.25713700797,
-      "required_pages_per_second_upper_estimate": 0.006496498328945833
+      "required_token_positions_per_second": 75739.11626551286,
+      "required_pages_per_second_upper_estimate": 0.006564874638859454
     },
     {
       "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": 9368.782142125996,
-      "required_pages_per_second_upper_estimate": 0.0037373472722698245
+      "required_token_positions_per_second": 9467.389533189107,
+      "required_pages_per_second_upper_estimate": 0.003776683234876778
     },
     {
       "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": 18737.564284251992,
-      "required_pages_per_second_upper_estimate": 0.004334605727293335
+      "required_token_positions_per_second": 18934.779066378214,
+      "required_pages_per_second_upper_estimate": 0.004380227896276616
     },
     {
       "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": 28106.346426377986,
-      "required_pages_per_second_upper_estimate": 0.0017115290347632305
+      "required_token_positions_per_second": 28402.16859956732,
+      "required_pages_per_second_upper_estimate": 0.0017295430530514678
     },
     {
       "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": 28106.346426377986,
-      "required_pages_per_second_upper_estimate": 0.002163704351249121
+      "required_token_positions_per_second": 28402.16859956732,
+      "required_pages_per_second_upper_estimate": 0.002186477561029423
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28106.346426377986,
-      "required_pages_per_second_upper_estimate": 0.34898675672520685
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28402.16859956732,
+      "required_pages_per_second_upper_estimate": 0.352659878063093
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06516983085933248,
-    "producer_p95_seconds": 0.22491542100033257,
-    "gpu_update_seconds": 0.9836924223652269,
-    "producer_batches_per_second": 15.344523482936086,
-    "gpu_batches_per_second": 1.0165779234077685,
-    "producer_headroom": 15.094291474969507,
+    "producer_batch_seconds": 0.12090066516398679,
+    "producer_p95_seconds": 0.2944767460012372,
+    "gpu_update_seconds": 0.9734467952007435,
+    "producer_batches_per_second": 8.271253087347565,
+    "gpu_batches_per_second": 1.0272775101116653,
+    "producer_headroom": 8.051624810172743,
     "sustainable": true,
-    "buffer_seconds": 60.00523776427884,
+    "buffer_seconds": 60.3537013024461,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 473314.50186431524,
-    "upload_bytes_per_second": 4054731.184984265,
+    "generation_bytes_per_second": 489904.4653120648,
+    "upload_bytes_per_second": 4069243.3801506446,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.56227390238851,
+    "estimated_full_shards_per_hour": 105.12209386369189,
     "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": 0.8450371549988631,
-      "bytes": 895470,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 32.774606008000774,
       "bytes": 10411787,
@@ -562,6 +556,12 @@
       "bytes": 10949848,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 0.9190237900002103,
+      "bytes": 2687904,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -155,3 +155,4 @@
 {"step": 330500, "loss": 0.5742582809180021, "lr": 0.0003, "grad_norm": 0.19433099567572845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48376117798, "tokens_per_second": 147441.66743168022, "data_wait_s": 1.3945998944109306e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 330600, "loss": 0.6046997979283333, "lr": 0.0003, "grad_norm": 0.19152793513634442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48390796240, "tokens_per_second": 149141.94140897636, "data_wait_s": 1.0780999218695797e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 330700, "loss": 0.5738974809646606, "lr": 0.0003, "grad_norm": 0.1722429909213363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48405610354, "tokens_per_second": 151068.1783950212, "data_wait_s": 1.1067999366787262e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 330800, "loss": 0.5701847933232784, "lr": 0.0003, "grad_norm": 0.15614280457714352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48420285230, "tokens_per_second": 147885.39125797685, "data_wait_s": 1.1126001481898129e-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 sha256:09beeeb0959c474e263b80d5ecaca20e4b54ceed913e2e6cc459bfddb2f0876a
-size 335040
+oid sha256:8b27bdee72c3554ce8b5a756ee9ea8cb4a67c1f37a1e2705d9edc064cb8a0847
+size 337158