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": 75543.33588885606,
-      "required_pages_per_second_upper_estimate": 0.006547904892011671
+      "required_token_positions_per_second": 73571.50457539439,
+      "required_pages_per_second_upper_estimate": 0.006376991551321588
     },
     {
       "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": 9442.916986107008,
-      "required_pages_per_second_upper_estimate": 0.003766920769948543
+      "required_token_positions_per_second": 9196.438071924298,
+      "required_pages_per_second_upper_estimate": 0.0036685966458928905
     },
     {
       "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": 18885.833972214015,
-      "required_pages_per_second_upper_estimate": 0.004368905310145965
+      "required_token_positions_per_second": 18392.876143848596,
+      "required_pages_per_second_upper_estimate": 0.00425486819231509
     },
     {
       "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": 28328.75095832102,
-      "required_pages_per_second_upper_estimate": 0.0017250723038921623
+      "required_token_positions_per_second": 27589.314215772894,
+      "required_pages_per_second_upper_estimate": 0.0016800444857957442
     },
     {
       "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": 28328.75095832102,
-      "required_pages_per_second_upper_estimate": 0.002180825667773244
+      "required_token_positions_per_second": 27589.314215772894,
+      "required_pages_per_second_upper_estimate": 0.0021239017804399792
     },
     {
       "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": 28328.75095832102,
-      "required_pages_per_second_upper_estimate": 0.35174827667185293
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27589.314215772894,
+      "required_pages_per_second_upper_estimate": 0.34256694706498747
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1121665883983951,
-    "producer_p95_seconds": 0.27618343999711215,
-    "gpu_update_seconds": 0.9759696091323411,
-    "producer_batches_per_second": 8.915310827215176,
-    "gpu_batches_per_second": 1.0246220688050138,
-    "producer_headroom": 8.701072423330524,
+    "producer_batch_seconds": 0.41821049948410405,
+    "producer_p95_seconds": 0.29720335899764905,
+    "gpu_update_seconds": 1.0021271200787425,
+    "producer_batches_per_second": 2.3911403497367467,
+    "gpu_batches_per_second": 0.9978773949570636,
+    "producer_headroom": 2.396226592385763,
     "sustainable": true,
-    "buffer_seconds": 60.510115766205146,
+    "buffer_seconds": 60.12762720472455,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360530.1354653012,
-    "upload_bytes_per_second": 3952974.4334514355,
+    "generation_bytes_per_second": 329836.168115089,
+    "upload_bytes_per_second": 4000300.810892834,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.36137435883786,
+    "estimated_full_shards_per_hour": 70.77516348447325,
     "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": 1.1375984709993645,
-      "bytes": 7150328,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 0.9804463260006742,
       "bytes": 5040957,
@@ -562,6 +556,12 @@
       "bytes": 5952533,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.4523514239990618,
+      "bytes": 13500548,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -244,3 +244,4 @@
 {"step": 139400, "loss": 0.6256307102739811, "lr": 0.0003, "grad_norm": 0.14989689171879073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20401218446, "tokens_per_second": 150023.24210697704, "data_wait_s": 1.4225999620975927e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 139500, "loss": 0.6047249548137188, "lr": 0.0003, "grad_norm": 0.13372338313865564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20415982038, "tokens_per_second": 150755.5205487166, "data_wait_s": 1.6589998267591e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 139600, "loss": 0.6191224306821823, "lr": 0.0003, "grad_norm": 0.14548936283670907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20430764940, "tokens_per_second": 147550.14833101322, "data_wait_s": 1.6154001059476286e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 139700, "loss": 0.6310807839035988, "lr": 0.0003, "grad_norm": 0.15356001626441015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20445353412, "tokens_per_second": 149751.25880204746, "data_wait_s": 1.5083998732734472e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50588fc80509e6aebbc06fe0d4ddc56fd8a3d944389ab2d36851312bd95e02d2
-size 524325
+oid sha256:24c61060cf0eb7e6abd3ddc4bb811ea39acbb6ccc717ed8b703f568ca11e2f3b
+size 526443