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": 76045.09419990663,
-      "required_pages_per_second_upper_estimate": 0.005260976065678003
+      "required_token_positions_per_second": 76850.69799816857,
+      "required_pages_per_second_upper_estimate": 0.005316709605700109
     },
     {
       "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": 9505.636774988328,
-      "required_pages_per_second_upper_estimate": 0.003979811617419534
+      "required_token_positions_per_second": 9606.337249771072,
+      "required_pages_per_second_upper_estimate": 0.00402197280334603
     },
     {
       "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": 19011.273549976657,
-      "required_pages_per_second_upper_estimate": 0.003959546919461558
+      "required_token_positions_per_second": 19212.674499542143,
+      "required_pages_per_second_upper_estimate": 0.004001493425955839
     },
     {
       "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": 28516.910324964985,
-      "required_pages_per_second_upper_estimate": 0.0017095452311807946
+      "required_token_positions_per_second": 28819.011749313213,
+      "required_pages_per_second_upper_estimate": 0.001727655750288305
     },
     {
       "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": 28516.910324964985,
-      "required_pages_per_second_upper_estimate": 0.0022170643451336373
+      "required_token_positions_per_second": 28819.011749313213,
+      "required_pages_per_second_upper_estimate": 0.0022405514020729786
     },
     {
       "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": 28516.910324964985,
-      "required_pages_per_second_upper_estimate": 0.365784306575916
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28819.011749313213,
+      "required_pages_per_second_upper_estimate": 0.36965933927621775
     }
   ],
   "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": 0.10320096055474437,
-    "producer_p95_seconds": 0.3114894509999431,
-    "gpu_update_seconds": 0.969529997637777,
-    "producer_batches_per_second": 9.689832290558344,
-    "gpu_batches_per_second": 1.0314276014527266,
-    "producer_headroom": 9.394583077775488,
+    "producer_batch_seconds": 0.08875697400804938,
+    "producer_p95_seconds": 0.3127830570010701,
+    "gpu_update_seconds": 0.9593666930879016,
+    "producer_batches_per_second": 11.266720290726786,
+    "gpu_batches_per_second": 1.0423543022755068,
+    "producer_headroom": 10.808916187260918,
     "sustainable": true,
-    "buffer_seconds": 60.11085985354217,
+    "buffer_seconds": 60.4401016645378,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288185.3457097499,
-    "upload_bytes_per_second": 5381869.818351038,
+    "generation_bytes_per_second": 287975.77452029206,
+    "upload_bytes_per_second": 5285715.605522252,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.837866577810026,
+    "estimated_full_shards_per_hour": 61.79289747911998,
     "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": 1.395833352999034,
-      "bytes": 13910666,
-      "pages": 50,
-      "shards": 4
-    },
     {
       "seconds": 0.8840833290014416,
       "bytes": 6069784,
@@ -562,6 +556,12 @@
       "bytes": 6773206,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.4528134669999417,
+      "bytes": 6484124,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -324,3 +324,4 @@
 {"step": 636800, "loss": 0.5594694018363953, "lr": 0.0001, "grad_norm": 0.16506547113416456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93210708814, "tokens_per_second": 152369.67231532015, "data_wait_s": 9.011000656755641e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 636900, "loss": 0.5780323147773743, "lr": 0.0001, "grad_norm": 0.1959813526404393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93225427449, "tokens_per_second": 152857.05821997774, "data_wait_s": 9.268002031603828e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 637000, "loss": 0.5176790952682495, "lr": 0.0001, "grad_norm": 0.17257702674561864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93240168892, "tokens_per_second": 152955.16928854142, "data_wait_s": 1.0569998266873881e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 637100, "loss": 0.5054295659065247, "lr": 0.0001, "grad_norm": 0.21669725697026537, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93254994326, "tokens_per_second": 153752.79937595213, "data_wait_s": 1.6162997781066224e-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 sha256:0f192b87fddf4ba11633e73fe168582538e066d32cb424d93da0daf93e048318
-size 696114
+oid sha256:d7564889c7e3f681f6630f0617fbb8505f52b70c72c9f56b12b18394812838f6
+size 698232