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": 75251.07172907058,
-      "required_pages_per_second_upper_estimate": 0.006522572175378207
+      "required_token_positions_per_second": 75063.3910963452,
+      "required_pages_per_second_upper_estimate": 0.006506304493805786
     },
     {
       "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": 9406.383966133822,
-      "required_pages_per_second_upper_estimate": 0.003752347202063915
+      "required_token_positions_per_second": 9382.92388704315,
+      "required_pages_per_second_upper_estimate": 0.0037429886257552064
     },
     {
       "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": 18812.767932267645,
-      "required_pages_per_second_upper_estimate": 0.004352002767722756
+      "required_token_positions_per_second": 18765.8477740863,
+      "required_pages_per_second_upper_estimate": 0.004341148614894082
     },
     {
       "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": 28219.151898401466,
-      "required_pages_per_second_upper_estimate": 0.001718398296164889
+      "required_token_positions_per_second": 28148.771661129453,
+      "required_pages_per_second_upper_estimate": 0.0017141125089715906
     },
     {
       "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": 28219.151898401466,
-      "required_pages_per_second_upper_estimate": 0.002172388428751021
+      "required_token_positions_per_second": 28148.771661129453,
+      "required_pages_per_second_upper_estimate": 0.0021669703632608557
     },
     {
       "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": 28219.151898401466,
-      "required_pages_per_second_upper_estimate": 0.3503874231521098
+      "required_token_positions_per_second": 28148.771661129453,
+      "required_pages_per_second_upper_estimate": 0.3495135361526932
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07747241943742722,
-    "producer_p95_seconds": 0.19249318599941034,
-    "gpu_update_seconds": 0.9797601323931419,
-    "producer_batches_per_second": 12.907819418337363,
-    "gpu_batches_per_second": 1.0206579824363955,
-    "producer_headroom": 12.646566862216982,
+    "producer_batch_seconds": 0.385722796414079,
+    "producer_p95_seconds": 0.20996136400026444,
+    "gpu_update_seconds": 0.9822098219006492,
+    "producer_batches_per_second": 2.5925353888767453,
+    "gpu_batches_per_second": 1.018112400937842,
+    "producer_headroom": 2.546413722579758,
     "sustainable": true,
-    "buffer_seconds": 60.745128208374794,
+    "buffer_seconds": 60.89700895784025,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300242.6435032342,
-    "upload_bytes_per_second": 5199527.108432241,
+    "generation_bytes_per_second": 319807.06828180427,
+    "upload_bytes_per_second": 5237485.30439386,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.42508200476426,
+    "estimated_full_shards_per_hour": 68.62315212574573,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 6141587,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1525904230002197,
+      "bytes": 7408784,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -63,3 +63,4 @@
 {"step": 201300, "loss": 0.5749208759516478, "lr": 0.0003, "grad_norm": 0.1506905842714099, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29462379900, "tokens_per_second": 147481.41415537766, "data_wait_s": 1.6646000403852668e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 201400, "loss": 0.5935119315981865, "lr": 0.0003, "grad_norm": 0.1516739405008383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29476965032, "tokens_per_second": 147825.77492817727, "data_wait_s": 1.715399957902264e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 201500, "loss": 0.5972956931218505, "lr": 0.0003, "grad_norm": 0.14020461508248058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29491660534, "tokens_per_second": 148147.6052728274, "data_wait_s": 1.3040999874647241e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 201600, "loss": 0.6152750365436077, "lr": 0.0003, "grad_norm": 0.1450862373672012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29506288345, "tokens_per_second": 148300.4665426418, "data_wait_s": 1.6713999684725422e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:803adc69f6d3540627aa0c51d5420a062e6c47b601f2d5648253ac1be977848d
-size 138618
+oid sha256:575040478c8cb7e784b64825676b756f8feabb6e9f6f39a7acd3f48af580969b
+size 140736