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": 74258.54275167065,
-      "required_pages_per_second_upper_estimate": 0.0064365422791588765
+      "required_token_positions_per_second": 75069.12627729523,
+      "required_pages_per_second_upper_estimate": 0.006506801604754851
     },
     {
       "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": 9282.317843958832,
-      "required_pages_per_second_upper_estimate": 0.003702855370974482
+      "required_token_positions_per_second": 9383.640784661904,
+      "required_pages_per_second_upper_estimate": 0.0037432746069339016
     },
     {
       "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": 18564.635687917664,
-      "required_pages_per_second_upper_estimate": 0.004294601739970737
+      "required_token_positions_per_second": 18767.281569323808,
+      "required_pages_per_second_upper_estimate": 0.0043414802981883455
     },
     {
       "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": 27846.953531876494,
-      "required_pages_per_second_upper_estimate": 0.001695733368417429
+      "required_token_positions_per_second": 28150.922353985712,
+      "required_pages_per_second_upper_estimate": 0.0017142434748827235
     },
     {
       "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": 27846.953531876494,
-      "required_pages_per_second_upper_estimate": 0.0021437355681849075
+      "required_token_positions_per_second": 28150.922353985712,
+      "required_pages_per_second_upper_estimate": 0.0021671359295504294
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27846.953531876494,
-      "required_pages_per_second_upper_estimate": 0.34576596510767094
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28150.922353985712,
+      "required_pages_per_second_upper_estimate": 0.349540240560062
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.384757290984453,
-    "producer_p95_seconds": 0.30032371100242017,
-    "gpu_update_seconds": 0.9928554650817099,
-    "producer_batches_per_second": 2.5990410667498107,
-    "gpu_batches_per_second": 1.0071959466101164,
-    "producer_headroom": 2.5804721270943465,
+    "producer_batch_seconds": 0.08239215942217015,
+    "producer_p95_seconds": 0.2779464840023138,
+    "gpu_update_seconds": 0.9821347823825564,
+    "producer_batches_per_second": 12.137077205078317,
+    "gpu_batches_per_second": 1.0181901893079324,
+    "producer_headroom": 11.920245679569877,
     "sustainable": true,
-    "buffer_seconds": 60.5641833699843,
+    "buffer_seconds": 60.892356507718496,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333884.9604340915,
-    "upload_bytes_per_second": 4007190.0885105394,
+    "generation_bytes_per_second": 334614.1456741783,
+    "upload_bytes_per_second": 3971091.9088957147,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.64394006506976,
+    "estimated_full_shards_per_hour": 71.80040624302876,
     "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": 1.3326636759993562,
-      "bytes": 12247274,
-      "pages": 47,
-      "shards": 3
-    },
     {
       "seconds": 1.6376108280001063,
       "bytes": 7445606,
@@ -562,6 +556,12 @@
       "bytes": 6073870,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0066101270022045,
+      "bytes": 7063346,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -136,3 +136,4 @@
 {"step": 293800, "loss": 0.5998360477387905, "lr": 0.0003, "grad_norm": 0.14458538680261215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43004171777, "tokens_per_second": 149164.57439300106, "data_wait_s": 8.318998879985884e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 293900, "loss": 0.6085012704133987, "lr": 0.0003, "grad_norm": 0.14893825711803013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43018855944, "tokens_per_second": 150892.0930995991, "data_wait_s": 1.2694999895757064e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 294000, "loss": 0.5516666574403644, "lr": 0.0003, "grad_norm": 0.14728797657168682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43033494776, "tokens_per_second": 149271.74959701372, "data_wait_s": 1.4050998288439587e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 294100, "loss": 0.5556954760104418, "lr": 0.0003, "grad_norm": 0.15022196021235193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43048083582, "tokens_per_second": 149508.01861430283, "data_wait_s": 1.106999843614176e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:80633ea0d5ecfa391b78622523600c5a64f4803bf00dea4c83aa5436e4230141
-size 294015
+oid sha256:2e24853b1c609377a01f2a9035c4802708b6ef7fa2923f09406e5f86db983a0d
+size 296133