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": 74222.010950236,
-      "required_pages_per_second_upper_estimate": 0.006433375795199524
+      "required_token_positions_per_second": 75187.25385051925,
+      "required_pages_per_second_upper_estimate": 0.006517040603410312
     },
     {
       "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": 9277.7513687795,
-      "required_pages_per_second_upper_estimate": 0.0037010337357505584
+      "required_token_positions_per_second": 9398.406731314906,
+      "required_pages_per_second_upper_estimate": 0.0037491649638243603
     },
     {
       "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": 18555.502737559,
-      "required_pages_per_second_upper_estimate": 0.004292488992639697
+      "required_token_positions_per_second": 18796.813462629812,
+      "required_pages_per_second_upper_estimate": 0.004348311981961119
     },
     {
       "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": 27833.2541063385,
-      "required_pages_per_second_upper_estimate": 0.0016948991452774961
+      "required_token_positions_per_second": 28195.22019394472,
+      "required_pages_per_second_upper_estimate": 0.0017169409809234225
     },
     {
       "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": 27833.2541063385,
-      "required_pages_per_second_upper_estimate": 0.0021426809484845558
+      "required_token_positions_per_second": 28195.22019394472,
+      "required_pages_per_second_upper_estimate": 0.002170546099894743
     },
     {
       "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": 27833.2541063385,
-      "required_pages_per_second_upper_estimate": 0.34559586409151694
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28195.22019394472,
+      "required_pages_per_second_upper_estimate": 0.3500902714770195
     }
   ],
   "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": 0.20855013382040966,
-    "producer_p95_seconds": 0.35660255100083305,
-    "gpu_update_seconds": 0.993344144898375,
-    "producer_batches_per_second": 4.79501010946911,
-    "gpu_batches_per_second": 1.0067004523415255,
-    "producer_headroom": 4.763095216969656,
+    "producer_batch_seconds": 0.1889642739297983,
+    "producer_p95_seconds": 0.3678609470007359,
+    "gpu_update_seconds": 0.9805917389479285,
+    "producer_batches_per_second": 5.2920056220336535,
+    "gpu_batches_per_second": 1.0197923970610792,
+    "producer_headroom": 5.189296995432194,
     "sustainable": true,
-    "buffer_seconds": 60.59399283880087,
+    "buffer_seconds": 60.796687814771566,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374795.5954121424,
+    "generation_bytes_per_second": 374494.94115030614,
     "upload_bytes_per_second": 4189189.617109923,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.42240998051838,
+    "estimated_full_shards_per_hour": 80.35789657480133,
     "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

@@ -309,3 +309,4 @@
 {"step": 345900, "loss": 0.6154396384954453, "lr": 0.0003, "grad_norm": 0.15911986876923162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50630570454, "tokens_per_second": 148408.74728900185, "data_wait_s": 1.6066001990111545e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 346000, "loss": 0.6023015528917313, "lr": 0.0003, "grad_norm": 0.1422033520210749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50645356753, "tokens_per_second": 150134.3683929201, "data_wait_s": 9.304996638093144e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 346100, "loss": 0.607675364241004, "lr": 0.0003, "grad_norm": 0.16171144785741007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50660075345, "tokens_per_second": 149965.76371127076, "data_wait_s": 1.42700009746477e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 346200, "loss": 0.5823972150683403, "lr": 0.0003, "grad_norm": 0.14689265943939842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50674792131, "tokens_per_second": 149689.65688336254, "data_wait_s": 1.903999873320572e-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 sha256:965e7be98b51d516cc5a530c241166290ed660ab4a2f22023b4c97ef1bab4743
-size 663561
+oid sha256:f80d509b5b8ef6060e1a25e59215c77b47202f6eb8f5bde7198548101c2f860f
+size 665679