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": 75706.47792407223,
-      "required_pages_per_second_upper_estimate": 0.006562045630144501
+      "required_token_positions_per_second": 73842.70465154399,
+      "required_pages_per_second_upper_estimate": 0.006400498486571926
     },
     {
       "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": 9463.309740509028,
-      "required_pages_per_second_upper_estimate": 0.0037750557445783584
+      "required_token_positions_per_second": 9230.338081442998,
+      "required_pages_per_second_upper_estimate": 0.00368211986654021
     },
     {
       "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": 18926.619481018057,
-      "required_pages_per_second_upper_estimate": 0.004378340319807341
+      "required_token_positions_per_second": 18460.676162885997,
+      "required_pages_per_second_upper_estimate": 0.004270552533479812
     },
     {
       "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": 28389.929221527087,
-      "required_pages_per_second_upper_estimate": 0.0017287977391438713
+      "required_token_positions_per_second": 27691.01424432899,
+      "required_pages_per_second_upper_estimate": 0.0016862374839559978
     },
     {
       "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": 28389.929221527087,
-      "required_pages_per_second_upper_estimate": 0.002185535339827119
+      "required_token_positions_per_second": 27691.01424432899,
+      "required_pages_per_second_upper_estimate": 0.002131730930161925
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28389.929221527087,
-      "required_pages_per_second_upper_estimate": 0.35250790595039655
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27691.01424432899,
+      "required_pages_per_second_upper_estimate": 0.34382972105155385
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10723449250011186,
-    "producer_p95_seconds": 0.33707956899888813,
-    "gpu_update_seconds": 0.9738664645572801,
-    "producer_batches_per_second": 9.325357696815294,
-    "gpu_batches_per_second": 1.026834824273983,
-    "producer_headroom": 9.081653130929531,
+    "producer_batch_seconds": 0.4214452942187279,
+    "producer_p95_seconds": 0.3454046920014662,
+    "gpu_update_seconds": 0.9984466352893592,
+    "producer_batches_per_second": 2.3727872008958895,
+    "gpu_batches_per_second": 1.0015557814065752,
+    "producer_headroom": 2.3691013969921575,
     "sustainable": true,
-    "buffer_seconds": 60.37972080255137,
+    "buffer_seconds": 60.90524475265091,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330535.91384688293,
+    "generation_bytes_per_second": 318802.38711100386,
     "upload_bytes_per_second": 2780524.4248856073,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.92531262926927,
+    "estimated_full_shards_per_hour": 68.40757093427264,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -270,3 +270,4 @@
 {"step": 222000, "loss": 0.6181822307407856, "lr": 0.0003, "grad_norm": 0.14457738280646545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32494021399, "tokens_per_second": 151195.13838724297, "data_wait_s": 8.519000402884558e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 222100, "loss": 0.5670797917991877, "lr": 0.0003, "grad_norm": 0.177642068664115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32508712610, "tokens_per_second": 149269.8434673818, "data_wait_s": 8.057999366428703e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 222200, "loss": 0.6142886988818645, "lr": 0.0003, "grad_norm": 0.13491388336804971, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32523294487, "tokens_per_second": 146836.3134184859, "data_wait_s": 2.0747000235132873e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 222300, "loss": 0.5637040808796883, "lr": 0.0003, "grad_norm": 0.1353372322068904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32537770694, "tokens_per_second": 141829.89712486847, "data_wait_s": 1.053700179909356e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ecf4e0796d622298ac5713cb85b0b159b78a917da8b26e344d1251429310ac35
-size 580176
+oid sha256:7044e2275bf1dc423b7a0eef49309bcba7e86a061873f412a9a314eee0c14dcd
+size 582294