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": 76893.43224476885,
-      "required_pages_per_second_upper_estimate": 0.00531966606003703
+      "required_token_positions_per_second": 77458.34544110506,
+      "required_pages_per_second_upper_estimate": 0.005358748065738775
     },
     {
       "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": 9611.679030596106,
-      "required_pages_per_second_upper_estimate": 0.004024209295428403
+      "required_token_positions_per_second": 9682.293180138133,
+      "required_pages_per_second_upper_estimate": 0.004053773965250526
     },
     {
       "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": 19223.35806119221,
-      "required_pages_per_second_upper_estimate": 0.004003718530103081
+      "required_token_positions_per_second": 19364.586360276266,
+      "required_pages_per_second_upper_estimate": 0.004033132660361585
     },
     {
       "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": 28835.03709178832,
-      "required_pages_per_second_upper_estimate": 0.0017286164451004065
+      "required_token_positions_per_second": 29046.8795404144,
+      "required_pages_per_second_upper_estimate": 0.0017413160764308503
     },
     {
       "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": 28835.03709178832,
-      "required_pages_per_second_upper_estimate": 0.002241797301962386
+      "required_token_positions_per_second": 29046.8795404144,
+      "required_pages_per_second_upper_estimate": 0.0022582671205466074
     },
     {
       "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": 28835.03709178832,
-      "required_pages_per_second_upper_estimate": 0.3698648951628163
+      "required_token_positions_per_second": 29046.8795404144,
+      "required_pages_per_second_upper_estimate": 0.3725821826350919
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5207096025312694,
-    "producer_p95_seconds": 0.2780029800014745,
-    "gpu_update_seconds": 0.9588335160447439,
-    "producer_batches_per_second": 1.9204562296120675,
-    "gpu_batches_per_second": 1.0429339225907233,
-    "producer_headroom": 1.8413977990489707,
+    "producer_batch_seconds": 0.13204462721887467,
+    "producer_p95_seconds": 0.271481791001861,
+    "gpu_update_seconds": 0.9518406258246065,
+    "producer_batches_per_second": 7.573197191449668,
+    "gpu_batches_per_second": 1.0505960481920718,
+    "producer_headroom": 7.2084767542026045,
     "sustainable": true,
-    "buffer_seconds": 60.406511510818866,
+    "buffer_seconds": 60.917800052774815,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 252334.91543122218,
-    "upload_bytes_per_second": 4732322.730198549,
+    "generation_bytes_per_second": 253477.79679820492,
+    "upload_bytes_per_second": 4760684.705199581,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.14519879534244,
+    "estimated_full_shards_per_hour": 54.39043453178034,
     "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": 1.3156736879973323,
-      "bytes": 6047068,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.5381603519999771,
       "bytes": 16328441,
@@ -562,6 +556,12 @@
       "bytes": 5725164,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 2.086732669002231,
+      "bytes": 12324271,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -287,3 +287,4 @@
 {"step": 913700, "loss": 0.5376759171485901, "lr": 3e-05, "grad_norm": 0.3368546632935128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133739144873, "tokens_per_second": 152205.26156926007, "data_wait_s": 1.5039000572869554e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 913800, "loss": 0.517048180103302, "lr": 3e-05, "grad_norm": 0.3136645450805417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133753869538, "tokens_per_second": 124523.07830525062, "data_wait_s": 1.5218000044114888e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 913900, "loss": 0.5273124575614929, "lr": 3e-05, "grad_norm": 0.3295266797638707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133768519253, "tokens_per_second": 153023.1055769892, "data_wait_s": 1.5828001778572798e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 914000, "loss": 0.49982258677482605, "lr": 3e-05, "grad_norm": 0.31729965561477075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133783250680, "tokens_per_second": 153413.05644599695, "data_wait_s": 1.2131997209507972e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:32fdde5fd802c0f11e165119e9c48bb1ab980fc95c59b11c8bb8c3b67664e798
-size 616182
+oid sha256:80204a4e927c372adeb4659d136de6d1e437811b684cf04d75fde282c3ddd10a
+size 618300