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": 76671.82599164997,
-      "required_pages_per_second_upper_estimate": 0.005304334825248916
+      "required_token_positions_per_second": 77201.16606681874,
+      "required_pages_per_second_upper_estimate": 0.005340955799887284
     },
     {
       "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": 9583.978248956246,
-      "required_pages_per_second_upper_estimate": 0.004012611556613893
+      "required_token_positions_per_second": 9650.145758352342,
+      "required_pages_per_second_upper_estimate": 0.004040314511063321
     },
     {
       "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": 19167.95649791249,
-      "required_pages_per_second_upper_estimate": 0.00399217984550939
+      "required_token_positions_per_second": 19300.291516704685,
+      "required_pages_per_second_upper_estimate": 0.004019741740014672
     },
     {
       "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": 28751.934746868734,
-      "required_pages_per_second_upper_estimate": 0.0017236345864124107
+      "required_token_positions_per_second": 28950.437275057026,
+      "required_pages_per_second_upper_estimate": 0.001735534509881488
     },
     {
       "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": 28751.934746868734,
-      "required_pages_per_second_upper_estimate": 0.002235336460173473
+      "required_token_positions_per_second": 28950.437275057026,
+      "required_pages_per_second_upper_estimate": 0.002250769158619766
     },
     {
       "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": 28751.934746868734,
-      "required_pages_per_second_upper_estimate": 0.3687989475105339
+      "required_token_positions_per_second": 28950.437275057026,
+      "required_pages_per_second_upper_estimate": 0.3713451248067242
     }
   ],
   "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": 0.3877824057967132,
-    "producer_p95_seconds": 0.24709614500170574,
-    "gpu_update_seconds": 0.9616048534963735,
-    "producer_batches_per_second": 2.5787657847587573,
-    "gpu_batches_per_second": 1.0399281954162591,
-    "producer_headroom": 2.4797536946544056,
+    "producer_batch_seconds": 0.08417656491386083,
+    "producer_p95_seconds": 0.24331038300078944,
+    "gpu_update_seconds": 0.9550114817720154,
+    "producer_batches_per_second": 11.879791020497395,
+    "gpu_batches_per_second": 1.0471078296823289,
+    "producer_headroom": 11.3453368256271,
     "sustainable": true,
-    "buffer_seconds": 60.58110577027153,
+    "buffer_seconds": 60.16572335163697,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353855.2482930009,
-    "upload_bytes_per_second": 5852950.981057861,
+    "generation_bytes_per_second": 338774.79068444687,
+    "upload_bytes_per_second": 5932337.949165163,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.92909895508309,
+    "estimated_full_shards_per_hour": 72.69318380737357,
     "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": 0.8772162379991642,
-      "bytes": 2610307,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.183507477000603,
       "bytes": 15672429,
@@ -562,6 +556,12 @@
       "bytes": 2540670,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1479466919990955,
+      "bytes": 10073767,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -151,3 +151,4 @@
 {"step": 500100, "loss": 0.4749933183193207, "lr": 0.0003, "grad_norm": 0.23643239184333206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73197547318, "tokens_per_second": 140835.37727070332, "data_wait_s": 1.2268003047211096e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 500200, "loss": 0.616510808467865, "lr": 0.0003, "grad_norm": 0.20496095867998576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73212342952, "tokens_per_second": 143988.05483712174, "data_wait_s": 1.4752000424778089e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 500300, "loss": 0.5731918811798096, "lr": 0.0003, "grad_norm": 0.1497603220181708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73227063816, "tokens_per_second": 153315.51502411708, "data_wait_s": 1.499900099588558e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 500400, "loss": 0.6094852089881897, "lr": 0.0003, "grad_norm": 0.15628840103694772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73241804958, "tokens_per_second": 153279.55766352548, "data_wait_s": 1.5287001588148996e-05, "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 sha256:66d3305080150fd80878647a4e4c8f2617ce7811a51aa1f72833732e2735f756
-size 326568
+oid sha256:f14c7503186e2271041d41945c75fd459fefae37b766a75fae844504c195e11c
+size 328686