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": 75420.38483050685,
-      "required_pages_per_second_upper_estimate": 0.006537247805890571
+      "required_token_positions_per_second": 75360.50547826392,
+      "required_pages_per_second_upper_estimate": 0.006532057615400984
     },
     {
       "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": 9427.548103813357,
-      "required_pages_per_second_upper_estimate": 0.003760789893016338
+      "required_token_positions_per_second": 9420.06318478299,
+      "required_pages_per_second_upper_estimate": 0.0037578040469056126
     },
     {
       "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": 18855.096207626713,
-      "required_pages_per_second_upper_estimate": 0.004361794669274873
+      "required_token_positions_per_second": 18840.12636956598,
+      "required_pages_per_second_upper_estimate": 0.004358331660699674
     },
     {
       "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": 28282.644311440068,
-      "required_pages_per_second_upper_estimate": 0.0017222646509999935
+      "required_token_positions_per_second": 28260.189554348974,
+      "required_pages_per_second_upper_estimate": 0.0017208972741041514
     },
     {
       "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": 28282.644311440068,
-      "required_pages_per_second_upper_estimate": 0.0021772762504649813
+      "required_token_positions_per_second": 28260.189554348974,
+      "required_pages_per_second_upper_estimate": 0.002175547621105363
     },
     {
       "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": 28282.644311440068,
-      "required_pages_per_second_upper_estimate": 0.35117578642661224
+      "required_token_positions_per_second": 28260.189554348974,
+      "required_pages_per_second_upper_estimate": 0.3508969734947785
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.1375645904921612,
+    "producer_batch_seconds": 0.1335078021796221,
     "producer_p95_seconds": 0.4377787170014926,
-    "gpu_update_seconds": 0.9775606444556049,
-    "producer_batches_per_second": 7.269312520193798,
-    "gpu_batches_per_second": 1.0229544383478033,
-    "producer_headroom": 7.106193831989846,
+    "gpu_update_seconds": 0.9783373868327517,
+    "producer_batches_per_second": 7.4901989522274866,
+    "gpu_batches_per_second": 1.0221422726544043,
+    "producer_headroom": 7.327941669779654,
     "sustainable": true,
-    "buffer_seconds": 60.6087599562475,
+    "buffer_seconds": 60.656917983630606,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276039.8165502997,
-    "upload_bytes_per_second": 5472745.03926124,
+    "generation_bytes_per_second": 294686.9514899694,
+    "upload_bytes_per_second": 5457329.4873138815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.231718753640585,
+    "estimated_full_shards_per_hour": 63.23295982860862,
     "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": 1.1231011929994565,
-      "bytes": 6897324,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0080412499955855,
       "bytes": 7466769,
@@ -562,6 +556,12 @@
       "bytes": 11946609,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.137292609004362,
+      "bytes": 5787517,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -409,3 +409,4 @@
 {"step": 355900, "loss": 0.5479207634925842, "lr": 0.0003, "grad_norm": 0.14617998126550985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52093511172, "tokens_per_second": 147155.4351942816, "data_wait_s": 1.212100323755294e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356000, "loss": 0.5907339053228498, "lr": 0.0003, "grad_norm": 0.16440081078503044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52108078318, "tokens_per_second": 145821.20666688864, "data_wait_s": 2.11730002774857e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356100, "loss": 0.5713396035134792, "lr": 0.0003, "grad_norm": 0.14518442927444247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52122825287, "tokens_per_second": 148726.7354203944, "data_wait_s": 1.1285999789834023e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 356200, "loss": 0.5745569840073586, "lr": 0.0003, "grad_norm": 0.13863634003555855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52137511384, "tokens_per_second": 148395.937101354, "data_wait_s": 1.2736003554891795e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:122355d49c87adcd6fd4329dd9eba1d69450d407a1b1b343186f4bcc181d5ccd
-size 876927
+oid sha256:431f4bc5e3b9dd0e95ca9f664b6c3f3c4d1b962a84b43d9b60fc6de45da5ec0a
+size 879045