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": 76887.78338546204,
-      "required_pages_per_second_upper_estimate": 0.005319275258843025
+      "required_token_positions_per_second": 75420.18282690577,
+      "required_pages_per_second_upper_estimate": 0.005217743246899648
     },
     {
       "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": 9610.972923182755,
-      "required_pages_per_second_upper_estimate": 0.004023913662999633
+      "required_token_positions_per_second": 9427.522853363222,
+      "required_pages_per_second_upper_estimate": 0.003947106949639275
     },
     {
       "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": 19221.94584636551,
-      "required_pages_per_second_upper_estimate": 0.0040034244029972885
+      "required_token_positions_per_second": 18855.045706726443,
+      "required_pages_per_second_upper_estimate": 0.003927008779717823
     },
     {
       "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": 28832.91876954826,
-      "required_pages_per_second_upper_estimate": 0.0017284894549166058
+      "required_token_positions_per_second": 28282.568560089665,
+      "required_pages_per_second_upper_estimate": 0.0016954968001956757
     },
     {
       "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": 28832.91876954826,
-      "required_pages_per_second_upper_estimate": 0.0022416326117257364
+      "required_token_positions_per_second": 28282.568560089665,
+      "required_pages_per_second_upper_estimate": 0.002198845303675071
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28832.91876954826,
-      "required_pages_per_second_upper_estimate": 0.36983772359959804
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28282.568560089665,
+      "required_pages_per_second_upper_estimate": 0.36277842203267874
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.43587853379690955,
-    "producer_p95_seconds": 0.2714741759991739,
-    "gpu_update_seconds": 0.9589039604689724,
-    "producer_batches_per_second": 2.294217132670116,
-    "gpu_batches_per_second": 1.042857305032851,
-    "producer_headroom": 2.199933894693144,
+    "producer_batch_seconds": 0.10948994828888203,
+    "producer_p95_seconds": 0.28354223299902515,
+    "gpu_update_seconds": 0.977563262730489,
+    "producer_batches_per_second": 9.133258492017603,
+    "gpu_batches_per_second": 1.0229516984986133,
+    "producer_headroom": 8.928337970817672,
     "sustainable": true,
-    "buffer_seconds": 60.41094950954526,
+    "buffer_seconds": 60.60892228929032,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314152.0210194562,
-    "upload_bytes_per_second": 5052860.958432231,
+    "generation_bytes_per_second": 290120.77171807125,
+    "upload_bytes_per_second": 5344833.395406782,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.40971062600865,
+    "estimated_full_shards_per_hour": 62.25316394478419,
     "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": 4.034771163000187,
-      "bytes": 5791973,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.1227300080008717,
       "bytes": 6451625,
@@ -562,6 +556,12 @@
       "bytes": 1167810,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.4214221570000518,
+      "bytes": 16238584,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -214,3 +214,4 @@
 {"step": 625800, "loss": 0.5401984453201294, "lr": 0.0001, "grad_norm": 0.16631244845804258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91600588830, "tokens_per_second": 150900.90455086165, "data_wait_s": 1.777400029823184e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 625900, "loss": 0.550552487373352, "lr": 0.0001, "grad_norm": 0.1606772406117453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91615241549, "tokens_per_second": 150707.62347697216, "data_wait_s": 1.7790996935218573e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 626000, "loss": 0.58170086145401, "lr": 0.0001, "grad_norm": 0.16856763967320154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91629823459, "tokens_per_second": 138877.91846774865, "data_wait_s": 1.8421000277157873e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 626100, "loss": 0.5501269698143005, "lr": 0.0001, "grad_norm": 0.15863774938725253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91644575305, "tokens_per_second": 151858.25455336555, "data_wait_s": 2.2718002583133057e-05, "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 sha256:7b9c5538d727e035ab2adf761b0cb61cb911db334c74731e69282b8a0ab92a1c
-size 461568
+oid sha256:a913cdba4a0ccb673ed5c390db130d63e9fa771903e34334c621087d0bca14a3
+size 463686