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": 77085.74469170555,
-      "required_pages_per_second_upper_estimate": 0.005332970681342465
+      "required_token_positions_per_second": 76935.27622299654,
+      "required_pages_per_second_upper_estimate": 0.0053225609235422276
     },
     {
       "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": 9635.718086463194,
-      "required_pages_per_second_upper_estimate": 0.004034273946127383
+      "required_token_positions_per_second": 9616.909527874568,
+      "required_pages_per_second_upper_estimate": 0.004026399195413692
     },
     {
       "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": 19271.436172926387,
-      "required_pages_per_second_upper_estimate": 0.004013731932872241
+      "required_token_positions_per_second": 19233.819055749136,
+      "required_pages_per_second_upper_estimate": 0.00400589727939431
     },
     {
       "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": 28907.154259389583,
-      "required_pages_per_second_upper_estimate": 0.0017329397591815633
+      "required_token_positions_per_second": 28850.728583623706,
+      "required_pages_per_second_upper_estimate": 0.0017295571260764172
     },
     {
       "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": 28907.154259389583,
-      "required_pages_per_second_upper_estimate": 0.0022474040945334875
+      "required_token_positions_per_second": 28850.728583623706,
+      "required_pages_per_second_upper_estimate": 0.002243017246433009
     },
     {
       "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": 28907.154259389583,
-      "required_pages_per_second_upper_estimate": 0.37078993675542365
+      "required_token_positions_per_second": 28850.728583623706,
+      "required_pages_per_second_upper_estimate": 0.3700661687718693
     }
   ],
   "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": 0.3749112745780394,
-    "producer_p95_seconds": 0.29182390499772737,
-    "gpu_update_seconds": 0.9564414314847133,
-    "producer_batches_per_second": 2.6672977523161836,
-    "gpu_batches_per_second": 1.0455423270901905,
-    "producer_headroom": 2.551114080421249,
+    "producer_batch_seconds": 0.10418299547663423,
+    "producer_p95_seconds": 0.3046345500006282,
+    "gpu_update_seconds": 0.9583120204351997,
+    "producer_batches_per_second": 9.598495372733607,
+    "gpu_batches_per_second": 1.0435014678683343,
+    "producer_headroom": 9.198353493782259,
     "sustainable": true,
-    "buffer_seconds": 60.25581018353694,
+    "buffer_seconds": 60.37365728741758,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 469509.4228374853,
-    "upload_bytes_per_second": 4874387.321583518,
+    "generation_bytes_per_second": 471695.90150523477,
+    "upload_bytes_per_second": 4813722.014608014,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.74579252093716,
+    "estimated_full_shards_per_hour": 101.21495994441779,
     "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": 1.268450359999406,
-      "bytes": 10380594,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1230385859998933,
       "bytes": 790073,
@@ -562,6 +556,12 @@
       "bytes": 11912121,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.4696884980003233,
+      "bytes": 5989675,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -326,3 +326,4 @@
 {"step": 792600, "loss": 0.5497496128082275, "lr": 3e-05, "grad_norm": 0.2706153066848277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116013934420, "tokens_per_second": 153520.20475070903, "data_wait_s": 1.176499790744856e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 792700, "loss": 0.5422593355178833, "lr": 3e-05, "grad_norm": 0.26376113343857827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116028712499, "tokens_per_second": 152664.09741162587, "data_wait_s": 1.152500044554472e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 792800, "loss": 0.5439079999923706, "lr": 3e-05, "grad_norm": 0.2599205656364801, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116043455045, "tokens_per_second": 153464.6243648686, "data_wait_s": 9.176001185551286e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 792900, "loss": 0.5092544555664062, "lr": 3e-05, "grad_norm": 0.2613729770209938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116058026192, "tokens_per_second": 151748.72570455764, "data_wait_s": 1.908899866975844e-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:a81085255f7084d30d5c6e2defe804b12f3ef0510f1bf648f89e8489f5fa189d
-size 699567
+oid sha256:4ebf971b4c5a30885ee6ce65836c5398a7424aa03346e0fdc3a52d609a04fbee
+size 701685