Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76000.60282146966,
-      "required_pages_per_second_upper_estimate": 0.006587539630798172
+      "required_token_positions_per_second": 76163.22641952123,
+      "required_pages_per_second_upper_estimate": 0.0066016354321115425
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9500.075352683707,
-      "required_pages_per_second_upper_estimate": 0.0037897220969697254
+      "required_token_positions_per_second": 9520.403302440154,
+      "required_pages_per_second_upper_estimate": 0.0037978312200575054
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19000.150705367414,
-      "required_pages_per_second_upper_estimate": 0.004395350474455216
+      "required_token_positions_per_second": 19040.80660488031,
+      "required_pages_per_second_upper_estimate": 0.004404755501288131
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28500.226058051125,
-      "required_pages_per_second_upper_estimate": 0.001735514237805407
+      "required_token_positions_per_second": 28561.20990732046,
+      "required_pages_per_second_upper_estimate": 0.00173922783426838
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28500.226058051125,
-      "required_pages_per_second_upper_estimate": 0.002194026295623924
+      "required_token_positions_per_second": 28561.20990732046,
+      "required_pages_per_second_upper_estimate": 0.002198721001154777
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28500.226058051125,
-      "required_pages_per_second_upper_estimate": 0.3538774235202593
+      "required_token_positions_per_second": 28561.20990732046,
+      "required_pages_per_second_upper_estimate": 0.3546346388283703
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.29562939585946424,
-    "producer_p95_seconds": 0.36044196000148077,
-    "gpu_update_seconds": 0.9700975684783955,
-    "producer_batches_per_second": 3.3826135492810674,
-    "gpu_batches_per_second": 1.0308241485116871,
-    "producer_headroom": 3.2814651792596385,
+    "producer_batch_seconds": 0.11922914655474415,
+    "producer_p95_seconds": 0.36585430199920665,
+    "gpu_update_seconds": 0.9680262177168343,
+    "producer_batches_per_second": 8.387210920283232,
+    "gpu_batches_per_second": 1.0330298722265792,
+    "producer_headroom": 8.119040064355106,
     "sustainable": true,
-    "buffer_seconds": 60.14604924566052,
+    "buffer_seconds": 60.01762549844373,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342772.87636278145,
-    "upload_bytes_per_second": 4770838.003791168,
+    "generation_bytes_per_second": 329812.6223522705,
+    "upload_bytes_per_second": 4845278.101215866,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.55107992327292,
+    "estimated_full_shards_per_hour": 70.77011111188972,
     "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": 1.0141177349978534,
-      "bytes": 2660846,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.5187353149995033,
       "bytes": 5879969,
@@ -562,6 +556,12 @@
       "bytes": 2529461,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.5562968439990073,
+      "bytes": 11835464,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -393,3 +393,4 @@
 {"step": 194300, "loss": 0.6588104479014874, "lr": 0.0003, "grad_norm": 0.15547511138598508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28437767900, "tokens_per_second": 151689.48828735013, "data_wait_s": 1.3245997251942754e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 194400, "loss": 0.5852263439446688, "lr": 0.0003, "grad_norm": 0.12636299922965405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28452450530, "tokens_per_second": 150746.44901064775, "data_wait_s": 1.185700239147991e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 194500, "loss": 0.5801782682538033, "lr": 0.0003, "grad_norm": 0.1276526351946504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28466968617, "tokens_per_second": 149404.84753841374, "data_wait_s": 1.2269003491383046e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 194600, "loss": 0.6777424439787865, "lr": 0.0003, "grad_norm": 0.17632340123455828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28481509759, "tokens_per_second": 150106.601301932, "data_wait_s": 8.789000276010484e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ca5eb42e26cddc7ca5151838f0373236aef167406a9b3d085efb1fe337c6abc6
-size 842256
+oid sha256:4e17804e35a98a360d6ad3a0c3f976c28217998234dfdbb303f942d5f203a891
+size 844374