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": 77264.36184777587,
-      "required_pages_per_second_upper_estimate": 0.005345327830648313
+      "required_token_positions_per_second": 76496.42976778543,
+      "required_pages_per_second_upper_estimate": 0.005292200507506658
     },
     {
       "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": 9658.045230971984,
-      "required_pages_per_second_upper_estimate": 0.004043621855289418
+      "required_token_positions_per_second": 9562.05372097318,
+      "required_pages_per_second_upper_estimate": 0.00400343221458359
     },
     {
       "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": 19316.090461943968,
-      "required_pages_per_second_upper_estimate": 0.0040230322436618105
+      "required_token_positions_per_second": 19124.10744194636,
+      "required_pages_per_second_upper_estimate": 0.003983047243529013
     },
     {
       "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": 28974.13569291595,
-      "required_pages_per_second_upper_estimate": 0.0017369551938467417
+      "required_token_positions_per_second": 28686.161162919536,
+      "required_pages_per_second_upper_estimate": 0.001719691560485104
     },
     {
       "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": 28974.13569291595,
-      "required_pages_per_second_upper_estimate": 0.002252611606369973
+      "required_token_positions_per_second": 28686.161162919536,
+      "required_pages_per_second_upper_estimate": 0.002230222853328844
     },
     {
       "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": 28974.13569291595,
-      "required_pages_per_second_upper_estimate": 0.37164910266563983
+      "required_token_positions_per_second": 28686.161162919536,
+      "required_pages_per_second_upper_estimate": 0.36795527459780575
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08702209400740912,
-    "producer_p95_seconds": 0.3165696170035517,
-    "gpu_update_seconds": 0.9542303623144768,
-    "producer_batches_per_second": 11.491334601932921,
-    "gpu_batches_per_second": 1.0479649773190085,
-    "producer_headroom": 10.965380380679335,
+    "producer_batch_seconds": 0.5684590364529072,
+    "producer_p95_seconds": 0.34685552900191396,
+    "gpu_update_seconds": 0.9638096865933568,
+    "producer_batches_per_second": 1.759141707447979,
+    "gpu_batches_per_second": 1.03754923187643,
+    "producer_headroom": 1.695477817728739,
     "sustainable": true,
-    "buffer_seconds": 60.11651282581204,
+    "buffer_seconds": 60.720010255381474,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 518758.37199746916,
-    "upload_bytes_per_second": 5054055.116011258,
+    "generation_bytes_per_second": 437678.7442441656,
+    "upload_bytes_per_second": 5049891.482521497,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.31347055380874,
+    "estimated_full_shards_per_hour": 93.91566987508513,
     "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": 1.1646632219999447,
-      "bytes": 931521,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 1.2082039209999493,
       "bytes": 10963858,
@@ -562,6 +556,12 @@
       "bytes": 6265583,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.4603665260001435,
+      "bytes": 2072280,
+      "pages": 20,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -341,3 +341,4 @@
 {"step": 638500, "loss": 0.521386444568634, "lr": 0.0001, "grad_norm": 0.16600122590102329, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93460130845, "tokens_per_second": 153391.30282328484, "data_wait_s": 9.71899862634018e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 638600, "loss": 0.5342638492584229, "lr": 0.0001, "grad_norm": 0.17172977449817925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93474774785, "tokens_per_second": 152039.84697366503, "data_wait_s": 1.740399602567777e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 638700, "loss": 0.5350292921066284, "lr": 0.0001, "grad_norm": 0.16551516570421757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93489464400, "tokens_per_second": 152998.01491213532, "data_wait_s": 1.2072996469214559e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 638800, "loss": 0.5327662229537964, "lr": 0.0001, "grad_norm": 0.1819359424993719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93504217745, "tokens_per_second": 136444.39995466507, "data_wait_s": 1.4217999705579132e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8d5bdb8aff07038605b031487eeb94043f7a341d81f0f657432d8e82837d52ab
-size 732120
+oid sha256:4b742637ffdbe7d4758ed570573225d728e45cbe0d6a2605000e1e8461502aef
+size 734238