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": 77480.64520397873,
-      "required_pages_per_second_upper_estimate": 0.005360290815076953
+      "required_token_positions_per_second": 77346.37298184689,
+      "required_pages_per_second_upper_estimate": 0.005351001551195393
     },
     {
       "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": 9685.08065049734,
-      "required_pages_per_second_upper_estimate": 0.004054941020880927
+      "required_token_positions_per_second": 9668.29662273086,
+      "required_pages_per_second_upper_estimate": 0.0040479138989454565
     },
     {
       "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": 19370.16130099468,
-      "required_pages_per_second_upper_estimate": 0.004034293773492158
+      "required_token_positions_per_second": 19336.59324546172,
+      "required_pages_per_second_upper_estimate": 0.004027302432773797
     },
     {
       "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": 29055.241951492022,
-      "required_pages_per_second_upper_estimate": 0.0017418173902062912
+      "required_token_positions_per_second": 29004.889868192582,
+      "required_pages_per_second_upper_estimate": 0.0017387988596956692
     },
     {
       "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": 29055.241951492022,
-      "required_pages_per_second_upper_estimate": 0.0022589172612255334
+      "required_token_positions_per_second": 29004.889868192582,
+      "required_pages_per_second_upper_estimate": 0.0022550026082244104
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29055.241951492022,
-      "required_pages_per_second_upper_estimate": 0.37268944666553816
+      "required_token_positions_per_second": 29004.889868192582,
+      "required_pages_per_second_upper_estimate": 0.3720435842048278
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.15346793452326324,
-    "producer_p95_seconds": 0.2961468840003363,
-    "gpu_update_seconds": 0.9515666758569271,
-    "producer_batches_per_second": 6.516019148276322,
-    "gpu_batches_per_second": 1.0508985080834787,
-    "producer_headroom": 6.200426680745385,
+    "producer_batch_seconds": 0.14222204915623138,
+    "producer_p95_seconds": 0.28581588599990937,
+    "gpu_update_seconds": 0.9532185823025455,
+    "producer_batches_per_second": 7.031258556129344,
+    "gpu_batches_per_second": 1.0490773245150673,
+    "producer_headroom": 6.702326312676257,
     "sustainable": true,
-    "buffer_seconds": 60.900267254843335,
+    "buffer_seconds": 60.052770685060366,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 486981.2899085499,
-    "upload_bytes_per_second": 4779478.68930813,
+    "generation_bytes_per_second": 480012.94715541706,
+    "upload_bytes_per_second": 4767325.286488481,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.4948484701383,
+    "estimated_full_shards_per_hour": 102.99960433003315,
     "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": 1.2346669860016846,
-      "bytes": 6391793,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.537398492997454,
       "bytes": 7291338,
@@ -562,6 +556,12 @@
       "bytes": 11121979,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.6951102359998913,
+      "bytes": 7495029,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -322,3 +322,4 @@
 {"step": 917200, "loss": 0.5379730463027954, "lr": 3e-05, "grad_norm": 0.33187450429568816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134251481257, "tokens_per_second": 152422.41773234293, "data_wait_s": 1.5644000086467713e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 917300, "loss": 0.5258959531784058, "lr": 3e-05, "grad_norm": 0.3281349656087223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134266121491, "tokens_per_second": 152749.22483755366, "data_wait_s": 1.2715001503238454e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 917400, "loss": 0.540526807308197, "lr": 3e-05, "grad_norm": 0.3360390075632434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134280874026, "tokens_per_second": 154069.3524968311, "data_wait_s": 1.3517001207219437e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 917500, "loss": 0.47982466220855713, "lr": 3e-05, "grad_norm": 0.3159958809459773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134295486274, "tokens_per_second": 152662.04811808284, "data_wait_s": 1.4806999388383701e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fc74ab2f0cce91e723c832540ea054c9b18ac325b34bad85f2a41d11ea66fbc0
-size 691095
+oid sha256:575da69517c69482c193b998f28db7e1b8ac3b0f5209266cb09e90a8c2f4683d
+size 693213