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": 77387.01236158158,
-      "required_pages_per_second_upper_estimate": 0.005353813077781792
+      "required_token_positions_per_second": 76363.72834847453,
+      "required_pages_per_second_upper_estimate": 0.005283019915409015
     },
     {
       "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": 9673.376545197698,
-      "required_pages_per_second_upper_estimate": 0.0040500407564014775
+      "required_token_positions_per_second": 9545.466043559316,
+      "required_pages_per_second_upper_estimate": 0.003996487300440499
     },
     {
       "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": 19346.753090395396,
-      "required_pages_per_second_upper_estimate": 0.004029418460540352
+      "required_token_positions_per_second": 19090.932087118632,
+      "required_pages_per_second_upper_estimate": 0.003976137692011338
     },
     {
       "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": 29020.129635593094,
-      "required_pages_per_second_upper_estimate": 0.0017397124604815524
+      "required_token_positions_per_second": 28636.39813067795,
+      "required_pages_per_second_upper_estimate": 0.0017167083426859716
     },
     {
       "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": 29020.129635593094,
-      "required_pages_per_second_upper_estimate": 0.0022561874331071476
+      "required_token_positions_per_second": 28636.39813067795,
+      "required_pages_per_second_upper_estimate": 0.002226353996456507
     },
     {
       "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": 29020.129635593094,
-      "required_pages_per_second_upper_estimate": 0.37223906357785425
+      "required_token_positions_per_second": 28636.39813067795,
+      "required_pages_per_second_upper_estimate": 0.3673169678515918
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08333567682029752,
-    "producer_p95_seconds": 0.30526041900156997,
-    "gpu_update_seconds": 0.9527180046118684,
-    "producer_batches_per_second": 11.999662547366947,
-    "gpu_batches_per_second": 1.0496285313799585,
-    "producer_headroom": 11.432294558143207,
+    "producer_batch_seconds": 0.8063564649999932,
+    "producer_p95_seconds": 0.30417103999934625,
+    "gpu_update_seconds": 0.965484551298402,
+    "producer_batches_per_second": 1.2401463166789497,
+    "gpu_batches_per_second": 1.0357493536848217,
+    "producer_headroom": 1.1973421101031416,
     "sustainable": true,
-    "buffer_seconds": 60.02123429054771,
+    "buffer_seconds": 60.82552673179933,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 251712.36065602352,
-    "upload_bytes_per_second": 4573220.829429412,
+    "generation_bytes_per_second": 226842.95097520304,
+    "upload_bytes_per_second": 4622279.467791692,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.011613032918255,
+    "estimated_full_shards_per_hour": 48.67521664564198,
     "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": 0.9429252529989753,
-      "bytes": 341818,
-      "pages": 9,
-      "shards": 1
-    },
     {
       "seconds": 1.6665152259993192,
       "bytes": 10630836,
@@ -562,6 +556,12 @@
       "bytes": 6331951,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.2861647640020237,
+      "bytes": 6463355,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -303,3 +303,4 @@
 {"step": 872800, "loss": 0.4873090386390686, "lr": 3e-05, "grad_norm": 0.3048071329696636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127753321336, "tokens_per_second": 152553.46068532305, "data_wait_s": 1.0973999451380223e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 872900, "loss": 0.5579450130462646, "lr": 3e-05, "grad_norm": 0.3107906100421599, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127768125526, "tokens_per_second": 154554.00396586288, "data_wait_s": 8.735001756576821e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 873000, "loss": 0.5453442335128784, "lr": 3e-05, "grad_norm": 0.3081546320966736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127782875800, "tokens_per_second": 142249.13116805596, "data_wait_s": 1.3732998922932893e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 873100, "loss": 0.5723668932914734, "lr": 3e-05, "grad_norm": 0.42357364625010396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127797571780, "tokens_per_second": 152783.4733376409, "data_wait_s": 9.731000318424776e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c5d5020348eff76d6a9485aabd761259d8c1dccf06211e6b8b1b4321367d61c3
-size 650853
+oid sha256:60666d097e15d117a33f8b27f04c84da9fcf615bbfa15c4aa3206ee5046ebeab
+size 652971