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": 74984.06259129161,
-      "required_pages_per_second_upper_estimate": 0.006499428500043999
+      "required_token_positions_per_second": 74893.80031258588,
+      "required_pages_per_second_upper_estimate": 0.0064916047944934396
     },
     {
       "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": 9373.007823911452,
-      "required_pages_per_second_upper_estimate": 0.003739032959913616
+      "required_token_positions_per_second": 9361.725039073235,
+      "required_pages_per_second_upper_estimate": 0.003734532088348985
     },
     {
       "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": 18746.015647822904,
-      "required_pages_per_second_upper_estimate": 0.004336560801516546
+      "required_token_positions_per_second": 18723.45007814647,
+      "required_pages_per_second_upper_estimate": 0.004331340654112366
     },
     {
       "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": 28119.023471734356,
-      "required_pages_per_second_upper_estimate": 0.0017123010003141168
+      "required_token_positions_per_second": 28085.175117219707,
+      "required_pages_per_second_upper_estimate": 0.001710239813112766
     },
     {
       "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": 28119.023471734356,
-      "required_pages_per_second_upper_estimate": 0.002164680265292962
+      "required_token_positions_per_second": 28085.175117219707,
+      "required_pages_per_second_upper_estimate": 0.0021620745252641824
     },
     {
       "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": 28119.023471734356,
-      "required_pages_per_second_upper_estimate": 0.34914416320119146
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28085.175117219707,
+      "required_pages_per_second_upper_estimate": 0.3487238799212748
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.09147726959358238,
-    "producer_p95_seconds": 0.2932236640008341,
-    "gpu_update_seconds": 0.9832489392027488,
-    "producer_batches_per_second": 10.931677393114446,
-    "gpu_batches_per_second": 1.017036439226503,
-    "producer_headroom": 10.74856020048645,
+    "producer_batch_seconds": 0.0793058158593567,
+    "producer_p95_seconds": 0.28030279200174846,
+    "gpu_update_seconds": 0.9844339543764616,
+    "producer_batches_per_second": 12.609415705065437,
+    "gpu_batches_per_second": 1.015812178718884,
+    "producer_headroom": 12.413136964914228,
     "sustainable": true,
-    "buffer_seconds": 60.961434230570426,
+    "buffer_seconds": 60.05047121696416,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287813.0280946757,
+    "generation_bytes_per_second": 287132.20739136037,
     "upload_bytes_per_second": 5306007.97869946,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.757975884725596,
+    "estimated_full_shards_per_hour": 61.61188761048897,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -238,3 +238,4 @@
 {"step": 338800, "loss": 0.6380332540720701, "lr": 0.0003, "grad_norm": 0.16096344868583667, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49591185901, "tokens_per_second": 148972.36315423885, "data_wait_s": 1.1410000297473744e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 338900, "loss": 0.5484571140259504, "lr": 0.0003, "grad_norm": 0.1442884114183441, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49605719347, "tokens_per_second": 147974.9637834689, "data_wait_s": 1.793300179997459e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 339000, "loss": 0.6315003428608179, "lr": 0.0003, "grad_norm": 0.1682675100232813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49620516109, "tokens_per_second": 149760.62995231096, "data_wait_s": 1.692300065769814e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 339100, "loss": 0.6156422272324562, "lr": 0.0003, "grad_norm": 0.16153957412485814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49635297006, "tokens_per_second": 150227.11843808705, "data_wait_s": 1.4793000445934013e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d57528eb0682fd534e60db0f47757cda0184c02d157ecfc44621908842689d2f
-size 511617
+oid sha256:713538d31eb8de6ecdbbde345b612bdbff22adb2e1f388403e2aacebedb51984
+size 513735