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": 75267.18871598822,
-      "required_pages_per_second_upper_estimate": 0.0065239691549560976
+      "required_token_positions_per_second": 75343.52361819918,
+      "required_pages_per_second_upper_estimate": 0.006530585670810703
     },
     {
       "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": 9408.398589498527,
-      "required_pages_per_second_upper_estimate": 0.003753150865445399
+      "required_token_positions_per_second": 9417.940452274897,
+      "required_pages_per_second_upper_estimate": 0.003756957257170455
     },
     {
       "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": 18816.797178997054,
-      "required_pages_per_second_upper_estimate": 0.004352934863041284
+      "required_token_positions_per_second": 18835.880904549795,
+      "required_pages_per_second_upper_estimate": 0.004357349547085812
     },
     {
       "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": 28225.195768495578,
-      "required_pages_per_second_upper_estimate": 0.0017187663361438847
+      "required_token_positions_per_second": 28253.82135682469,
+      "required_pages_per_second_upper_estimate": 0.0017205094842862726
     },
     {
       "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": 28225.195768495578,
-      "required_pages_per_second_upper_estimate": 0.002172853702601369
+      "required_token_positions_per_second": 28253.82135682469,
+      "required_pages_per_second_upper_estimate": 0.0021750573796316445
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28225.195768495578,
-      "required_pages_per_second_upper_estimate": 0.35046246779114665
+      "required_token_positions_per_second": 28253.82135682469,
+      "required_pages_per_second_upper_estimate": 0.35081790179451294
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.23451052994539623,
-    "producer_p95_seconds": 0.32018725800298853,
-    "gpu_update_seconds": 0.9795503360462132,
-    "producer_batches_per_second": 4.2642008451937805,
-    "gpu_batches_per_second": 1.020876583061906,
-    "producer_headroom": 4.176999370878114,
+    "producer_batch_seconds": 0.07346469244541254,
+    "producer_p95_seconds": 0.2693540540021786,
+    "gpu_update_seconds": 0.9785578966762188,
+    "producer_batches_per_second": 13.611981030793038,
+    "gpu_batches_per_second": 1.0219119414360782,
+    "producer_headroom": 13.320111527089423,
     "sustainable": true,
-    "buffer_seconds": 60.73212083486522,
+    "buffer_seconds": 60.67058959392557,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422712.8181698127,
-    "upload_bytes_per_second": 5636778.74826536,
+    "generation_bytes_per_second": 506242.1327536779,
+    "upload_bytes_per_second": 5744352.4235935975,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.7043305284575,
+    "estimated_full_shards_per_hour": 108.62777697522881,
     "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": 0.9184139579992916,
-      "bytes": 2547467,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.207064377000279,
       "bytes": 6456430,
@@ -562,6 +556,12 @@
       "bytes": 6457509,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.143103587000951,
+      "bytes": 11836705,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -193,3 +193,4 @@
 {"step": 134300, "loss": 0.6199606731534004, "lr": 0.0003, "grad_norm": 0.1368644183500872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19654668601, "tokens_per_second": 148750.87462226464, "data_wait_s": 1.5558001905446872e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 134400, "loss": 0.5725448876619339, "lr": 0.0003, "grad_norm": 0.1396398416571278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19669231759, "tokens_per_second": 147809.68250423536, "data_wait_s": 1.3234999642008916e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 134500, "loss": 0.6046467572450638, "lr": 0.0003, "grad_norm": 0.14199222364862882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19683785624, "tokens_per_second": 148626.3800544342, "data_wait_s": 1.4899000234436244e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 134600, "loss": 0.6050839927047491, "lr": 0.0003, "grad_norm": 0.1471052218670691, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19698342227, "tokens_per_second": 148477.49888373193, "data_wait_s": 1.4622000890085474e-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 sha256:1181b146d419c60b01c00c48519e4a485f00d47cd2b5525740b7b7facb5e161d
-size 415524
+oid sha256:a70a1ac099bffbe85a58a5bd4f912f0c14e616e5ebec16dd02b2830d067de88f
+size 417642