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": 76480.85816313203,
-      "required_pages_per_second_upper_estimate": 0.0052911232277656345
+      "required_token_positions_per_second": 76929.38189693759,
+      "required_pages_per_second_upper_estimate": 0.005322153140388748
     },
     {
       "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": 9560.107270391503,
-      "required_pages_per_second_upper_estimate": 0.004002617276371552
+      "required_token_positions_per_second": 9616.172737117198,
+      "required_pages_per_second_upper_estimate": 0.004026090716509521
     },
     {
       "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": 19120.214540783007,
-      "required_pages_per_second_upper_estimate": 0.003982236454879395
+      "required_token_positions_per_second": 19232.345474234397,
+      "required_pages_per_second_upper_estimate": 0.004005590371225748
     },
     {
       "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": 28680.32181117451,
-      "required_pages_per_second_upper_estimate": 0.0017193415002641635
+      "required_token_positions_per_second": 28848.518211351595,
+      "required_pages_per_second_upper_estimate": 0.0017294246176337471
     },
     {
       "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": 28680.32181117451,
-      "required_pages_per_second_upper_estimate": 0.002229768869415262
+      "required_token_positions_per_second": 28848.518211351595,
+      "required_pages_per_second_upper_estimate": 0.0022428453997112526
     },
     {
       "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": 28680.32181117451,
-      "required_pages_per_second_upper_estimate": 0.36788037366342796
+      "required_token_positions_per_second": 28848.518211351595,
+      "required_pages_per_second_upper_estimate": 0.37003781648967554
     }
   ],
   "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": 0.0579066022811503,
-    "producer_p95_seconds": 0.20035370100049477,
-    "gpu_update_seconds": 0.9640059195300837,
-    "producer_batches_per_second": 17.269187978682684,
-    "gpu_batches_per_second": 1.0373380284713003,
-    "producer_headroom": 16.64759943692787,
+    "producer_batch_seconds": 0.07536701057807704,
+    "producer_p95_seconds": 0.22511880899946846,
+    "gpu_update_seconds": 0.9583854462625674,
+    "producer_batches_per_second": 13.268404734775068,
+    "gpu_batches_per_second": 1.043421520954557,
+    "producer_headroom": 12.716245992929764,
     "sustainable": true,
-    "buffer_seconds": 60.73237293039528,
+    "buffer_seconds": 60.378283114541745,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 471403.28294959926,
-    "upload_bytes_per_second": 5108438.940122447,
+    "generation_bytes_per_second": 481362.92371220415,
+    "upload_bytes_per_second": 5099300.679883408,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.15217081418976,
+    "estimated_full_shards_per_hour": 103.28927787327379,
     "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": 1.0410187759998735,
-      "bytes": 7475284,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.175038504999975,
       "bytes": 877584,
@@ -562,6 +556,12 @@
       "bytes": 6227206,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9700920809991658,
+      "bytes": 6347868,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -152,3 +152,4 @@
 {"step": 619600, "loss": 0.5462628602981567, "lr": 0.0001, "grad_norm": 0.17916029485821122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90692360414, "tokens_per_second": 154038.50334648308, "data_wait_s": 1.5442001313203946e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 619700, "loss": 0.5706902742385864, "lr": 0.0001, "grad_norm": 0.1793077210142055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90707010731, "tokens_per_second": 152362.96802740998, "data_wait_s": 1.2308999430388212e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 619800, "loss": 0.5233141183853149, "lr": 0.0001, "grad_norm": 0.14545157516544222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90721640713, "tokens_per_second": 152442.2338918091, "data_wait_s": 1.400600012857467e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 619900, "loss": 0.5285357236862183, "lr": 0.0001, "grad_norm": 0.15875105633434458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90736332835, "tokens_per_second": 152907.69725344583, "data_wait_s": 1.3785998817184009e-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 sha256:46382bae090998e0ea382d2ee48921545ccd5bfa155ca74926fcda41011616e6
-size 328686
+oid sha256:37e96143cfd3266eb8a004a4c5dfaece74ce28a4fce1c2f7599f6231aa680cc4
+size 330804