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": 75723.90677631988,
-      "required_pages_per_second_upper_estimate": 0.005238755574456772
+      "required_token_positions_per_second": 76258.62531268995,
+      "required_pages_per_second_upper_estimate": 0.005275748643520796
     },
     {
       "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": 9465.488347039985,
-      "required_pages_per_second_upper_estimate": 0.003963002309032074
+      "required_token_positions_per_second": 9532.328164086244,
+      "required_pages_per_second_upper_estimate": 0.003990986744655245
     },
     {
       "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": 18930.97669407997,
-      "required_pages_per_second_upper_estimate": 0.003942823201948768
+      "required_token_positions_per_second": 19064.65632817249,
+      "required_pages_per_second_upper_estimate": 0.003970665144361218
     },
     {
       "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": 28396.465041119955,
-      "required_pages_per_second_upper_estimate": 0.0017023247203235847
+      "required_token_positions_per_second": 28596.984492258736,
+      "required_pages_per_second_upper_estimate": 0.001714345555243881
     },
     {
       "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": 28396.465041119955,
-      "required_pages_per_second_upper_estimate": 0.002207700254097501
+      "required_token_positions_per_second": 28596.984492258736,
+      "required_pages_per_second_upper_estimate": 0.002223289759431687
     },
     {
       "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": 28396.465041119955,
-      "required_pages_per_second_upper_estimate": 0.3642393637988219
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28596.984492258736,
+      "required_pages_per_second_upper_estimate": 0.36681141201701795
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12785764674217148,
-    "producer_p95_seconds": 0.2839835779996065,
-    "gpu_update_seconds": 0.9736423163926873,
-    "producer_batches_per_second": 7.821198226935366,
-    "gpu_batches_per_second": 1.027071218211804,
-    "producer_headroom": 7.615049558639728,
+    "producer_batch_seconds": 0.10205526479691684,
+    "producer_p95_seconds": 0.2866812979991664,
+    "gpu_update_seconds": 0.9668152251327189,
+    "producer_batches_per_second": 9.798612565357928,
+    "gpu_batches_per_second": 1.0343238025267194,
+    "producer_headroom": 9.473447813364814,
     "sustainable": true,
-    "buffer_seconds": 60.36582361634661,
+    "buffer_seconds": 60.90935918336129,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343621.8919241051,
-    "upload_bytes_per_second": 4876486.128122283,
+    "generation_bytes_per_second": 364236.79064211866,
+    "upload_bytes_per_second": 4870983.946111981,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.73325889866223,
+    "estimated_full_shards_per_hour": 78.1567362732665,
     "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": 1.0818150289996993,
-      "bytes": 2336300,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.2755690529993444,
       "bytes": 16950690,
@@ -562,6 +556,12 @@
       "bytes": 6223313,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.2432226130003983,
+      "bytes": 2637387,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -305,3 +305,4 @@
 {"step": 790500, "loss": 0.5351627469062805, "lr": 3e-05, "grad_norm": 0.26463694846194785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115706737281, "tokens_per_second": 149942.46834166712, "data_wait_s": 1.6002002666937187e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 790600, "loss": 0.59759122133255, "lr": 3e-05, "grad_norm": 0.3038910553041401, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115721224740, "tokens_per_second": 150316.62702318624, "data_wait_s": 1.3362001482164487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 790700, "loss": 0.5069711804389954, "lr": 3e-05, "grad_norm": 0.2505566734304559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115735872939, "tokens_per_second": 152687.89437593988, "data_wait_s": 9.497998689766973e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 790800, "loss": 0.534726083278656, "lr": 3e-05, "grad_norm": 0.2632059355800747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115750515759, "tokens_per_second": 151996.63088258673, "data_wait_s": 1.0162999387830496e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:54c4062db29812acfc8a82de5aaa381d565c370f3aabdeeafd60175c84a2b37f
-size 655089
+oid sha256:70e00650b2506bfb855205ab0ec78ca2e4ced5c748a6293a1bdea0805908777e
+size 657207