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": 74806.3840653734,
-      "required_pages_per_second_upper_estimate": 0.006484027775739514
+      "required_token_positions_per_second": 74777.04045882421,
+      "required_pages_per_second_upper_estimate": 0.006481484346294522
     },
     {
       "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": 9350.798008171674,
-      "required_pages_per_second_upper_estimate": 0.0037301731323486813
+      "required_token_positions_per_second": 9347.130057353026,
+      "required_pages_per_second_upper_estimate": 0.003728709931926371
     },
     {
       "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": 18701.59601634335,
-      "required_pages_per_second_upper_estimate": 0.004326285101532565
+      "required_token_positions_per_second": 18694.260114706052,
+      "required_pages_per_second_upper_estimate": 0.004324588069796236
     },
     {
       "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": 28052.394024515022,
-      "required_pages_per_second_upper_estimate": 0.0017082436165561522
+      "required_token_positions_per_second": 28041.39017205908,
+      "required_pages_per_second_upper_estimate": 0.0017075735396743373
     },
     {
       "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": 28052.394024515022,
-      "required_pages_per_second_upper_estimate": 0.0021595509460039033
+      "required_token_positions_per_second": 28041.39017205908,
+      "required_pages_per_second_upper_estimate": 0.002158703838981232
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28052.394024515022,
-      "required_pages_per_second_upper_estimate": 0.34831684846114236
+      "required_token_positions_per_second": 28041.39017205908,
+      "required_pages_per_second_upper_estimate": 0.3481802174411647
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.18034966557813448,
-    "producer_p95_seconds": 0.20051170500028093,
-    "gpu_update_seconds": 0.9855843310855529,
-    "producer_batches_per_second": 5.544784332115998,
-    "gpu_batches_per_second": 1.0146265199839057,
-    "producer_headroom": 5.4648525569822,
+    "producer_batch_seconds": 0.15772256171875654,
+    "producer_p95_seconds": 0.19882969300033437,
+    "gpu_update_seconds": 0.985971088821015,
+    "producer_batches_per_second": 6.340247007800653,
+    "gpu_batches_per_second": 1.0142285218482017,
+    "producer_headroom": 6.2513002456753926,
     "sustainable": true,
-    "buffer_seconds": 60.12064419621873,
+    "buffer_seconds": 60.14423641808192,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331610.51938075095,
+    "generation_bytes_per_second": 332557.9793351844,
     "upload_bytes_per_second": 5888761.815966367,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.15589796130081,
+    "estimated_full_shards_per_hour": 71.3592008117833,
     "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": [

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -49,3 +49,4 @@
 {"step": 119900, "loss": 0.574279710650444, "lr": 0.0003, "grad_norm": 0.1383865787803532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17545806061, "tokens_per_second": 148313.0207369725, "data_wait_s": 1.1406999874452595e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 120000, "loss": 0.6221815645694733, "lr": 0.0003, "grad_norm": 0.14926640171740102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17560585532, "tokens_per_second": 150396.44749049205, "data_wait_s": 1.4455999917117879e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 120100, "loss": 0.5857874490320683, "lr": 0.0003, "grad_norm": 0.1262709847639684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17575224687, "tokens_per_second": 125561.53278647548, "data_wait_s": 1.048400008585304e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 120200, "loss": 0.6386179234832525, "lr": 0.0003, "grad_norm": 0.15743646713979917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17589976380, "tokens_per_second": 150474.45886600437, "data_wait_s": 1.1791000360972248e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:017bd8c5d4526f61b5843fad7c3005411d983a7cbdd41b78c683f2b01f4cd8cd
-size 108966
+oid sha256:e8e5376d2d5af9f5e84f3695638730fe2c5530504d593cf2c524523ec2c609b5
+size 111084