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": 77364.14604702007,
-      "required_pages_per_second_upper_estimate": 0.005352231133083267
+      "required_token_positions_per_second": 76762.49207477359,
+      "required_pages_per_second_upper_estimate": 0.005310607315253704
     },
     {
       "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": 9670.518255877509,
-      "required_pages_per_second_upper_estimate": 0.004048844050350982
+      "required_token_positions_per_second": 9595.311509346699,
+      "required_pages_per_second_upper_estimate": 0.004017356556073987
     },
     {
       "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": 19341.036511755017,
-      "required_pages_per_second_upper_estimate": 0.004028227847965879
+      "required_token_positions_per_second": 19190.623018693397,
+      "required_pages_per_second_upper_estimate": 0.003996900684031702
     },
     {
       "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": 29011.554767632526,
-      "required_pages_per_second_upper_estimate": 0.0017391984102403818
+      "required_token_positions_per_second": 28785.934528040096,
+      "required_pages_per_second_upper_estimate": 0.001725672821378974
     },
     {
       "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": 29011.554767632526,
-      "required_pages_per_second_upper_estimate": 0.0022555207748400715
+      "required_token_positions_per_second": 28785.934528040096,
+      "required_pages_per_second_upper_estimate": 0.002237979793610314
     },
     {
       "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": 29011.554767632526,
-      "required_pages_per_second_upper_estimate": 0.37212907437863196
+      "required_token_positions_per_second": 28785.934528040096,
+      "required_pages_per_second_upper_estimate": 0.36923506019727936
     }
   ],
   "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": 0.41469655451565757,
-    "producer_p95_seconds": 0.19895376800013764,
-    "gpu_update_seconds": 0.9529995969346045,
-    "producer_batches_per_second": 2.4114017565637704,
-    "gpu_batches_per_second": 1.0493183871394867,
-    "producer_headroom": 2.2980649020526704,
+    "producer_batch_seconds": 0.19020915717956655,
+    "producer_p95_seconds": 0.2164476719990489,
+    "gpu_update_seconds": 0.9604690781558041,
+    "producer_batches_per_second": 5.257370438038123,
+    "gpu_batches_per_second": 1.0411579328718206,
+    "producer_headroom": 5.049541738146052,
     "sustainable": true,
-    "buffer_seconds": 60.03897460688008,
+    "buffer_seconds": 60.50955192381566,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314015.8654652145,
-    "upload_bytes_per_second": 4252136.214812416,
+    "generation_bytes_per_second": 355765.915302258,
+    "upload_bytes_per_second": 4277368.175340949,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.38049481360746,
+    "estimated_full_shards_per_hour": 76.33908361721807,
     "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": 1.650117919000877,
-      "bytes": 6106251,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0884742970001753,
       "bytes": 6215302,
@@ -562,6 +556,12 @@
       "bytes": 5834009,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.3078732329995546,
+      "bytes": 7188440,
+      "pages": 42,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -145,3 +145,4 @@
 {"step": 699500, "loss": 0.5255951285362244, "lr": 8e-05, "grad_norm": 0.20401954144534976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102387638859, "tokens_per_second": 153350.35853250782, "data_wait_s": 1.0024999937741086e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 699600, "loss": 0.5263522863388062, "lr": 8e-05, "grad_norm": 0.2090223796641753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102402160775, "tokens_per_second": 151392.39442331032, "data_wait_s": 1.4368999472935684e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 699700, "loss": 0.5336575508117676, "lr": 8e-05, "grad_norm": 0.20244858323315756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102416811239, "tokens_per_second": 152592.7532016044, "data_wait_s": 1.0017000022344291e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 699800, "loss": 0.5427844524383545, "lr": 8e-05, "grad_norm": 0.2044449421545245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102431389470, "tokens_per_second": 151725.389850195, "data_wait_s": 1.3274999218992889e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d427bde785fb3eeacf54cec8e3d7d8ea5b44a3f0620c26db4198db3f80f0f962
-size 313077
+oid sha256:a43cfd675ffb9f0275656339796a89c0a5e1c0f901da5f4b1d217727e1e1f07d
+size 315195