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": 75420.18282690577,
-      "required_pages_per_second_upper_estimate": 0.005217743246899648
+      "required_token_positions_per_second": 76401.54698455284,
+      "required_pages_per_second_upper_estimate": 0.005285636296404238
     },
     {
       "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": 9427.522853363222,
-      "required_pages_per_second_upper_estimate": 0.003947106949639275
+      "required_token_positions_per_second": 9550.193373069105,
+      "required_pages_per_second_upper_estimate": 0.003998466534588382
     },
     {
       "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": 18855.045706726443,
-      "required_pages_per_second_upper_estimate": 0.003927008779717823
+      "required_token_positions_per_second": 19100.38674613821,
+      "required_pages_per_second_upper_estimate": 0.0039781068481489915
     },
     {
       "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": 28282.568560089665,
-      "required_pages_per_second_upper_estimate": 0.0016954968001956757
+      "required_token_positions_per_second": 28650.580119207312,
+      "required_pages_per_second_upper_estimate": 0.0017175585312436642
     },
     {
       "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": 28282.568560089665,
-      "required_pages_per_second_upper_estimate": 0.002198845303675071
+      "required_token_positions_per_second": 28650.580119207312,
+      "required_pages_per_second_upper_estimate": 0.00222745658368469
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28282.568560089665,
-      "required_pages_per_second_upper_estimate": 0.36277842203267874
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28650.580119207312,
+      "required_pages_per_second_upper_estimate": 0.3674988791730136
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "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,
+    "producer_batch_seconds": 0.09374914150009772,
+    "producer_p95_seconds": 0.28397713499725796,
+    "gpu_update_seconds": 0.9650066380842605,
+    "producer_batches_per_second": 10.66676434577225,
+    "gpu_batches_per_second": 1.0362623017653108,
+    "producer_headroom": 10.293498400550735,
     "sustainable": true,
-    "buffer_seconds": 60.60892228929032,
+    "buffer_seconds": 60.79541819930841,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290120.77171807125,
-    "upload_bytes_per_second": 5344833.395406782,
+    "generation_bytes_per_second": 290430.43474539876,
+    "upload_bytes_per_second": 5295745.941802656,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.25316394478419,
+    "estimated_full_shards_per_hour": 62.319610421862336,
     "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": 1.1227300080008717,
-      "bytes": 6451625,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.4268143489989598,
       "bytes": 2671684,
@@ -562,6 +556,12 @@
       "bytes": 16238584,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.1309305639988452,
+      "bytes": 2518678,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -215,3 +215,4 @@
 {"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}
+{"step": 626200, "loss": 0.5858222246170044, "lr": 0.0001, "grad_norm": 0.16567491129721723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91659315953, "tokens_per_second": 152543.8206018042, "data_wait_s": 1.3728000340051949e-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:a913cdba4a0ccb673ed5c390db130d63e9fa771903e34334c621087d0bca14a3
-size 463686
+oid sha256:dd7e1a35897ed021dbabb792a3300f09abf5e8e2276770957686c8b88f4735c8
+size 465804