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": 76785.57944056058,
-      "required_pages_per_second_upper_estimate": 0.005312204552788918
+      "required_token_positions_per_second": 76081.1847816869,
+      "required_pages_per_second_upper_estimate": 0.005263472895867248
     },
     {
       "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": 9598.197430070073,
-      "required_pages_per_second_upper_estimate": 0.004018564830815986
+      "required_token_positions_per_second": 9510.148097710862,
+      "required_pages_per_second_upper_estimate": 0.003981700414036327
     },
     {
       "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": 19196.394860140146,
-      "required_pages_per_second_upper_estimate": 0.003998102806391365
+      "required_token_positions_per_second": 19020.296195421724,
+      "required_pages_per_second_upper_estimate": 0.003961426098564604
     },
     {
       "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": 28794.59229021022,
-      "required_pages_per_second_upper_estimate": 0.0017261918410014358
+      "required_token_positions_per_second": 28530.444293132587,
+      "required_pages_per_second_upper_estimate": 0.0017103565719072695
     },
     {
       "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": 28794.59229021022,
-      "required_pages_per_second_upper_estimate": 0.00223865289653757
+      "required_token_positions_per_second": 28530.444293132587,
+      "required_pages_per_second_upper_estimate": 0.0022181165516290337
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28794.59229021022,
-      "required_pages_per_second_upper_estimate": 0.36934611267441697
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28530.444293132587,
+      "required_pages_per_second_upper_estimate": 0.36595790578792714
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12360020544551276,
-    "producer_p95_seconds": 0.281286548000935,
-    "gpu_update_seconds": 0.9601802908457903,
-    "producer_batches_per_second": 8.090601438691253,
-    "gpu_batches_per_second": 1.041471075311423,
-    "producer_headroom": 7.768436042519937,
+    "producer_batch_seconds": 0.4321026642421941,
+    "producer_p95_seconds": 0.2735124450009607,
+    "gpu_update_seconds": 0.9690700823279855,
+    "producer_batches_per_second": 2.314264832765527,
+    "gpu_batches_per_second": 1.0319171112967516,
+    "producer_headroom": 2.2426848120168508,
     "sustainable": true,
-    "buffer_seconds": 60.491358323284786,
+    "buffer_seconds": 60.082345104335104,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384152.2702961368,
+    "generation_bytes_per_second": 344755.98417331884,
     "upload_bytes_per_second": 5207063.29804664,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.43013459837988,
+    "estimated_full_shards_per_hour": 73.9766086950271,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -228,3 +228,4 @@
 {"step": 627200, "loss": 0.5442319512367249, "lr": 0.0001, "grad_norm": 0.1723691441592288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91805318673, "tokens_per_second": 151585.0667025374, "data_wait_s": 1.603900091140531e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 627300, "loss": 0.5515968799591064, "lr": 0.0001, "grad_norm": 0.1660719319697747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91819960425, "tokens_per_second": 151817.48481567745, "data_wait_s": 1.1936001101275906e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 627400, "loss": 0.5578224062919617, "lr": 0.0001, "grad_norm": 0.18071927340125216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91834666325, "tokens_per_second": 151007.69225771772, "data_wait_s": 1.364099807688035e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 627500, "loss": 0.5221641063690186, "lr": 0.0001, "grad_norm": 0.15815645773418807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91849452402, "tokens_per_second": 152911.36024273763, "data_wait_s": 1.022600190481171e-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:d215f24cd07d6c5e3312957e67c92e83b780a8486f4b7f83166d58ccee3a3652
-size 491220
+oid sha256:adf03d9c21836c8b96ff1d8ca7d7e99ee88f4e5b6451e8ef52a24555eccf600f
+size 493338