Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74853.53869103121,
-      "required_pages_per_second_upper_estimate": 0.0051785414907843065
+      "required_token_positions_per_second": 76812.17456450872,
+      "required_pages_per_second_upper_estimate": 0.00531404446517284
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9356.692336378901,
-      "required_pages_per_second_upper_estimate": 0.00391745169128733
+      "required_token_positions_per_second": 9601.52182056359,
+      "required_pages_per_second_upper_estimate": 0.004019956683694454
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18713.384672757802,
-      "required_pages_per_second_upper_estimate": 0.0038975045221948464
+      "required_token_positions_per_second": 19203.04364112718,
+      "required_pages_per_second_upper_estimate": 0.003999487572130819
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28070.077009136705,
-      "required_pages_per_second_upper_estimate": 0.0016827582561718575
+      "required_token_positions_per_second": 28804.565461690774,
+      "required_pages_per_second_upper_estimate": 0.0017267897174035437
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28070.077009136705,
-      "required_pages_per_second_upper_estimate": 0.0021823250202400338
+      "required_token_positions_per_second": 28804.565461690774,
+      "required_pages_per_second_upper_estimate": 0.002239428266752838
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28070.077009136705,
-      "required_pages_per_second_upper_estimate": 0.36005280857270566
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28804.565461690774,
+      "required_pages_per_second_upper_estimate": 0.36947403780981225
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3831788458203107,
-    "producer_p95_seconds": 0.2198908680002205,
-    "gpu_update_seconds": 0.9849634538231113,
-    "producer_batches_per_second": 2.609747408835152,
-    "gpu_batches_per_second": 1.0152660955272246,
-    "producer_headroom": 2.5705058214121865,
+    "producer_batch_seconds": 0.09203949624217955,
+    "producer_p95_seconds": 0.22887637599887967,
+    "gpu_update_seconds": 0.9598478420641696,
+    "producer_batches_per_second": 10.864900839621537,
+    "gpu_batches_per_second": 1.0418317947660147,
+    "producer_headroom": 10.428651625151918,
     "sustainable": true,
-    "buffer_seconds": 60.082770683209795,
+    "buffer_seconds": 60.470414050042685,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287967.58479055605,
-    "upload_bytes_per_second": 4805509.442904458,
+    "generation_bytes_per_second": 299997.606441248,
+    "upload_bytes_per_second": 4754863.559105689,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.79114015376579,
+    "estimated_full_shards_per_hour": 64.37250275543289,
     "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": 1.4518775689994072,
-      "bytes": 12256015,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.9289161779997812,
       "bytes": 848234,
@@ -562,6 +556,12 @@
       "bytes": 7431338,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.9902176969990251,
+      "bytes": 5564439,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -161,3 +161,4 @@
 {"step": 736100, "loss": 0.5431311130523682, "lr": 5e-05, "grad_norm": 0.24596208616680798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107744761546, "tokens_per_second": 149158.29127311741, "data_wait_s": 1.95359989447752e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 736200, "loss": 0.5547208189964294, "lr": 5e-05, "grad_norm": 0.22673007027056988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107759247246, "tokens_per_second": 146478.99680238252, "data_wait_s": 1.1204001566511579e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 736300, "loss": 0.5168012380599976, "lr": 5e-05, "grad_norm": 0.213197359884553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107773787864, "tokens_per_second": 150692.86581329023, "data_wait_s": 9.845000022323802e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 736400, "loss": 0.5322527289390564, "lr": 5e-05, "grad_norm": 0.22228331554590866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107788455239, "tokens_per_second": 139548.78594190927, "data_wait_s": 1.5736000932520255e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9293f7c8e2c93c75e01820c167dc662535c6f2798f21fdfc17326ef81b1524a
-size 347748
+oid sha256:1955f44cd5b3a73e2000ecb98164bfca2843dfe16d7a657bda73d8f9fd528874
+size 349866