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": 75120.39307719535,
-      "required_pages_per_second_upper_estimate": 0.006511245281035676
+      "required_token_positions_per_second": 76329.52372956047,
+      "required_pages_per_second_upper_estimate": 0.006616049661469065
     },
     {
       "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": 9390.049134649418,
-      "required_pages_per_second_upper_estimate": 0.0037458309935572915
+      "required_token_positions_per_second": 9541.19046619506,
+      "required_pages_per_second_upper_estimate": 0.0038061235304751315
     },
     {
       "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": 18780.098269298836,
-      "required_pages_per_second_upper_estimate": 0.0043444452161613524
+      "required_token_positions_per_second": 19082.38093239012,
+      "required_pages_per_second_upper_estimate": 0.004414372990274351
     },
     {
       "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": 28170.147403948253,
-      "required_pages_per_second_upper_estimate": 0.0017154141795594005
+      "required_token_positions_per_second": 28623.571398585176,
+      "required_pages_per_second_upper_estimate": 0.0017430253219009428
     },
     {
       "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": 28170.147403948253,
-      "required_pages_per_second_upper_estimate": 0.00216861592711489
+      "required_token_positions_per_second": 28623.571398585176,
+      "required_pages_per_second_upper_estimate": 0.002203521761379991
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28170.147403948253,
-      "required_pages_per_second_upper_estimate": 0.3497789513384935
+      "required_token_positions_per_second": 28623.571398585176,
+      "required_pages_per_second_upper_estimate": 0.3554089598393928
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11613393781237846,
-    "producer_p95_seconds": 0.3885753840004327,
-    "gpu_update_seconds": 0.981464512895127,
-    "producer_batches_per_second": 8.610747373567593,
-    "gpu_batches_per_second": 1.018885539784008,
-    "producer_headroom": 8.451142976661512,
+    "producer_batch_seconds": 0.16264687696889268,
+    "producer_p95_seconds": 0.3858460460032802,
+    "gpu_update_seconds": 0.9659172021198795,
+    "producer_batches_per_second": 6.148288972011782,
+    "gpu_batches_per_second": 1.0352854238492901,
+    "producer_headroom": 5.938738081670131,
     "sustainable": true,
-    "buffer_seconds": 60.85079979949788,
+    "buffer_seconds": 60.85278373355241,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295215.5503309981,
+    "generation_bytes_per_second": 282437.18975245336,
     "upload_bytes_per_second": 5461863.241892408,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.34638483474213,
+    "estimated_full_shards_per_hour": 60.60444611959649,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -354,3 +354,4 @@
 {"step": 150400, "loss": 0.5901217982172966, "lr": 0.0003, "grad_norm": 0.13779970108432205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22010611914, "tokens_per_second": 145315.96758618526, "data_wait_s": 1.0153999028261751e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 150500, "loss": 0.6060477271676064, "lr": 0.0003, "grad_norm": 0.13552522216887578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22025162642, "tokens_per_second": 148150.53943653347, "data_wait_s": 8.285998774226755e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 150600, "loss": 0.6056232452392578, "lr": 0.0003, "grad_norm": 0.13553303769077266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22039723020, "tokens_per_second": 149647.2460471847, "data_wait_s": 1.2923002941533923e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 150700, "loss": 0.5843577682971954, "lr": 0.0003, "grad_norm": 0.14000978061212047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22054300582, "tokens_per_second": 149173.14354409723, "data_wait_s": 1.2389995390549302e-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 sha256:742158a93ad78feff7656b2ef9cf19b1c2a2553630c35880bdbbc7aaff06dec7
-size 759654
+oid sha256:7fa09c3c5fdd20c7111dbcec040c28d41f6b5555c499abebc27bb2c08b63d01c
+size 761772