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": 75922.93218729971,
-      "required_pages_per_second_upper_estimate": 0.006580807337082744
+      "required_token_positions_per_second": 75880.29483090602,
+      "required_pages_per_second_upper_estimate": 0.006577111639093931
     },
     {
       "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": 9490.366523412464,
-      "required_pages_per_second_upper_estimate": 0.003785849099813493
+      "required_token_positions_per_second": 9485.036853863252,
+      "required_pages_per_second_upper_estimate": 0.0037837230149446515
     },
     {
       "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": 18980.733046824927,
-      "required_pages_per_second_upper_estimate": 0.0043908585408905295
+      "required_token_positions_per_second": 18970.073707726504,
+      "required_pages_per_second_upper_estimate": 0.00438839269038807
     },
     {
       "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": 28471.099570237388,
-      "required_pages_per_second_upper_estimate": 0.0017337405875124218
+      "required_token_positions_per_second": 28455.110561589754,
+      "required_pages_per_second_upper_estimate": 0.001732766940773626
     },
     {
       "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": 28471.099570237388,
-      "required_pages_per_second_upper_estimate": 0.002191784057964744
+      "required_token_positions_per_second": 28455.110561589754,
+      "required_pages_per_second_upper_estimate": 0.0021905531798186405
     },
     {
       "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": 28471.099570237388,
-      "required_pages_per_second_upper_estimate": 0.3535157700216967
+      "required_token_positions_per_second": 28455.110561589754,
+      "required_pages_per_second_upper_estimate": 0.35331724004606274
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.11411454092188933,
-    "producer_p95_seconds": 0.2691598500023247,
-    "gpu_update_seconds": 0.9710899971317642,
-    "producer_batches_per_second": 8.76312511903714,
-    "gpu_batches_per_second": 1.0297706731133314,
-    "producer_headroom": 8.509783146711067,
+    "producer_batch_seconds": 0.08543809809373215,
+    "producer_p95_seconds": 0.2401865150022786,
+    "gpu_update_seconds": 0.9716356554003611,
+    "producer_batches_per_second": 11.704380391320546,
+    "gpu_batches_per_second": 1.029192366955648,
+    "producer_headroom": 11.372393312575875,
     "sustainable": true,
-    "buffer_seconds": 60.207579822169386,
+    "buffer_seconds": 60.24141063482239,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441606.12367167394,
-    "upload_bytes_per_second": 5831001.606646914,
+    "generation_bytes_per_second": 437912.0583173251,
+    "upload_bytes_per_second": 5825708.135421485,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.75839407551445,
+    "estimated_full_shards_per_hour": 93.96573364391149,
     "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": 1.1212218569999095,
-      "bytes": 7621728,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0421719010009838,
       "bytes": 6283978,
@@ -562,6 +556,12 @@
       "bytes": 8178789,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.9424325320032949,
+      "bytes": 6190430,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -188,3 +188,4 @@
 {"step": 373800, "loss": 0.5403846129775047, "lr": 0.0003, "grad_norm": 0.2319163103574092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54713423612, "tokens_per_second": 148937.27212668382, "data_wait_s": 9.621999197406694e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 373900, "loss": 0.5923236533999443, "lr": 0.0003, "grad_norm": 0.16723194428624796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54727941165, "tokens_per_second": 148546.19985122856, "data_wait_s": 1.1596999684115872e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 374000, "loss": 0.5675775744020939, "lr": 0.0003, "grad_norm": 0.16214766662555807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54742535343, "tokens_per_second": 149163.61020735494, "data_wait_s": 1.4774999726796523e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 374100, "loss": 0.6167305372655392, "lr": 0.0003, "grad_norm": 0.15267573844008275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54757211934, "tokens_per_second": 149487.78632816183, "data_wait_s": 8.283997885882854e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c9ba50e314c51d651081222a3d5cf69a3ef2a59c0db36fa28825c1728a114ba
-size 404934
+oid sha256:05818db65283628410644ce8f2009e19d54dd06d6e9b809f276114f62d37f62b
+size 407052