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": 74184.8981630521,
-      "required_pages_per_second_upper_estimate": 0.0064301589528679235
+      "required_token_positions_per_second": 75049.83355513688,
+      "required_pages_per_second_upper_estimate": 0.006505129360495123
     },
     {
       "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": 9273.112270381513,
-      "required_pages_per_second_upper_estimate": 0.0036991831300388993
+      "required_token_positions_per_second": 9381.22919439211,
+      "required_pages_per_second_upper_estimate": 0.003742312587518793
     },
     {
       "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": 18546.224540763025,
-      "required_pages_per_second_upper_estimate": 0.004290342645101634
+      "required_token_positions_per_second": 18762.45838878422,
+      "required_pages_per_second_upper_estimate": 0.00434036454025559
     },
     {
       "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": 27819.336811144538,
-      "required_pages_per_second_upper_estimate": 0.0016940516550185887
+      "required_token_positions_per_second": 28143.68758317633,
+      "required_pages_per_second_upper_estimate": 0.001713802915298341
     },
     {
       "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": 27819.336811144538,
-      "required_pages_per_second_upper_estimate": 0.002141609556574986
+      "required_token_positions_per_second": 28143.68758317633,
+      "required_pages_per_second_upper_estimate": 0.0021665789768663957
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27819.336811144538,
-      "required_pages_per_second_upper_estimate": 0.34542305786339866
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28143.68758317633,
+      "required_pages_per_second_upper_estimate": 0.34945040891982976
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.13739134358573324,
-    "producer_p95_seconds": 0.3298838349983271,
-    "gpu_update_seconds": 0.9938410893003063,
-    "producer_batches_per_second": 7.278478933980236,
-    "gpu_batches_per_second": 1.006197077949383,
-    "producer_headroom": 7.23365143219625,
+    "producer_batch_seconds": 0.07964708617168981,
+    "producer_p95_seconds": 0.27390125400052057,
+    "gpu_update_seconds": 0.9823872553406029,
+    "producer_batches_per_second": 12.555387121687891,
+    "gpu_batches_per_second": 1.017928515016505,
+    "producer_headroom": 12.334252294213721,
     "sustainable": true,
-    "buffer_seconds": 60.62430644731869,
+    "buffer_seconds": 60.90800983111738,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303632.00522760756,
+    "generation_bytes_per_second": 303723.5439144328,
     "upload_bytes_per_second": 5074897.442387634,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.15236013051195,
+    "estimated_full_shards_per_hour": 65.17200220179308,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 260100, "loss": 0.518187377601862, "lr": 0.0003, "grad_norm": 0.15154002333609332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38071381151, "tokens_per_second": 134372.34560540706, "data_wait_s": 1.1162999726366252e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 260200, "loss": 0.5818884521722794, "lr": 0.0003, "grad_norm": 0.14185731552824035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38086085883, "tokens_per_second": 149267.4067283347, "data_wait_s": 1.50500018207822e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 260300, "loss": 0.5661483407020569, "lr": 0.0003, "grad_norm": 0.14240667021617626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38100805365, "tokens_per_second": 149176.6743370543, "data_wait_s": 1.5752000763313845e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 260400, "loss": 0.549375843256712, "lr": 0.0003, "grad_norm": 0.1408540863274199, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38115529406, "tokens_per_second": 150087.3628222345, "data_wait_s": 9.696999768493697e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a3865d571190a41f35676f1f773b5dc9bbc28015da996047eb97ca44ee5e89e9
-size 495456
+oid sha256:ea62b6eb264ea161117dff5bb6da157c45c8f40cceff378e203db5672fd7d89a
+size 497574