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": 75126.20852568044,
-      "required_pages_per_second_upper_estimate": 0.006511749349371245
+      "required_token_positions_per_second": 72914.42520475831,
+      "required_pages_per_second_upper_estimate": 0.006320037576827292
     },
     {
       "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": 9390.776065710055,
-      "required_pages_per_second_upper_estimate": 0.0037461209772259676
+      "required_token_positions_per_second": 9114.30315059479,
+      "required_pages_per_second_upper_estimate": 0.003635831797748041
     },
     {
       "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": 18781.55213142011,
-      "required_pages_per_second_upper_estimate": 0.0043447815415760095
+      "required_token_positions_per_second": 18228.60630118958,
+      "required_pages_per_second_upper_estimate": 0.004216867255266416
     },
     {
       "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": 28172.328197130166,
-      "required_pages_per_second_upper_estimate": 0.0017155469784224415
+      "required_token_positions_per_second": 27342.90945178437,
+      "required_pages_per_second_upper_estimate": 0.001665039728454728
     },
     {
       "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": 28172.328197130166,
-      "required_pages_per_second_upper_estimate": 0.0021687838106108554
+      "required_token_positions_per_second": 27342.90945178437,
+      "required_pages_per_second_upper_estimate": 0.0021049328596161
     },
     {
       "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": 28172.328197130166,
-      "required_pages_per_second_upper_estimate": 0.3498060294911676
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27342.90945178437,
+      "required_pages_per_second_upper_estimate": 0.339507424559946
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.10853620750765458,
-    "producer_p95_seconds": 0.42890520799846854,
-    "gpu_update_seconds": 0.9813885386588824,
-    "producer_batches_per_second": 9.213515221908546,
-    "gpu_batches_per_second": 1.0189644168522194,
-    "producer_headroom": 9.042038239540195,
+    "producer_batch_seconds": 0.4990021154372357,
+    "producer_p95_seconds": 0.4724845999953686,
+    "gpu_update_seconds": 1.0111579401875144,
+    "producer_batches_per_second": 2.0039995203703294,
+    "gpu_batches_per_second": 0.9889651856114138,
+    "producer_headroom": 2.026360027154429,
     "sustainable": true,
-    "buffer_seconds": 60.84608939685071,
+    "buffer_seconds": 60.66947641125086,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269967.7498793661,
-    "upload_bytes_per_second": 5527129.434853146,
+    "generation_bytes_per_second": 279593.68647207227,
+    "upload_bytes_per_second": 5391479.023901175,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.92879459653604,
+    "estimated_full_shards_per_hour": 59.994296508995305,
     "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": 1.474424235006154,
-      "bytes": 14927789,
-      "pages": 21,
-      "shards": 4
-    },
     {
       "seconds": 1.0885418239995488,
       "bytes": 8199005,
@@ -562,6 +556,12 @@
       "bytes": 7260405,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 0.8715928729943698,
+      "bytes": 1215608,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -417,3 +417,4 @@
 {"step": 356700, "loss": 0.6333474293351173, "lr": 0.0003, "grad_norm": 0.18712339975237002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52210646056, "tokens_per_second": 146762.10956158003, "data_wait_s": 1.3723998563364148e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356800, "loss": 0.5970645230263472, "lr": 0.0003, "grad_norm": 0.17870772044610614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52225208874, "tokens_per_second": 146988.22783714926, "data_wait_s": 1.4484001439996064e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356900, "loss": 0.615576034411788, "lr": 0.0003, "grad_norm": 0.15941947816925933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52239903919, "tokens_per_second": 140987.3160980433, "data_wait_s": 1.2231997970957309e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 357000, "loss": 0.604896292090416, "lr": 0.0003, "grad_norm": 0.15028696367232258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52254457624, "tokens_per_second": 145493.48765809817, "data_wait_s": 1.3530996511690319e-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 sha256:24302e6da5ebe148e20c91b610b13c460d0e5fa2c047fb6cfd97a0eb8f85ee03
-size 893871
+oid sha256:0e5449fe767cdef7df313592a756c27951154a6c6c6c286f9c44d867c1fe335d
+size 895989