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": 76131.3579957731,
-      "required_pages_per_second_upper_estimate": 0.006598873157910835
+      "required_token_positions_per_second": 75797.44255041578,
+      "required_pages_per_second_upper_estimate": 0.006569930213408222
     },
     {
       "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": 9516.419749471637,
-      "required_pages_per_second_upper_estimate": 0.0037962421212189394
+      "required_token_positions_per_second": 9474.680318801973,
+      "required_pages_per_second_upper_estimate": 0.003779591638264709
     },
     {
       "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": 19032.839498943275,
-      "required_pages_per_second_upper_estimate": 0.004402912451545873
+      "required_token_positions_per_second": 18949.360637603946,
+      "required_pages_per_second_upper_estimate": 0.0043836010861528
     },
     {
       "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": 28549.259248414914,
-      "required_pages_per_second_upper_estimate": 0.0017385001018412937
+      "required_token_positions_per_second": 28424.040956405915,
+      "required_pages_per_second_upper_estimate": 0.0017308749648275506
     },
     {
       "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": 28549.259248414914,
-      "required_pages_per_second_upper_estimate": 0.002197801006350687
+      "required_token_positions_per_second": 28424.040956405915,
+      "required_pages_per_second_upper_estimate": 0.0021881613556054167
     },
     {
       "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": 28549.259248414914,
-      "required_pages_per_second_upper_estimate": 0.3544862516410459
+      "required_token_positions_per_second": 28424.040956405915,
+      "required_pages_per_second_upper_estimate": 0.35293145953295896
     }
   ],
   "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": 0.1340564343671531,
-    "producer_p95_seconds": 0.3527957210026216,
-    "gpu_update_seconds": 0.9684314314226926,
-    "producer_batches_per_second": 7.459544964930255,
-    "gpu_batches_per_second": 1.0325976290659329,
-    "producer_headroom": 7.224057808149346,
+    "producer_batch_seconds": 0.08307194946860363,
+    "producer_p95_seconds": 0.2995998249971308,
+    "gpu_update_seconds": 0.9726977259286906,
+    "producer_batches_per_second": 12.037757707587467,
+    "gpu_batches_per_second": 1.0280686109811168,
+    "producer_headroom": 11.709099547450897,
     "sustainable": true,
-    "buffer_seconds": 60.04274874820694,
+    "buffer_seconds": 60.30725900757882,
     "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": 401458.3315093647,
-    "upload_bytes_per_second": 4637308.27795431,
+    "generation_bytes_per_second": 440483.9708887232,
+    "upload_bytes_per_second": 4566402.222583162,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.1436124702521,
+    "estimated_full_shards_per_hour": 94.51760621067307,
     "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": 1.1401566910026304,
-      "bytes": 8280901,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.1491751960020338,
       "bytes": 6275126,
@@ -562,6 +556,12 @@
       "bytes": 7334524,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.0401804770008312,
+      "bytes": 1254870,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -328,3 +328,4 @@
 {"step": 187800, "loss": 0.6040479615330696, "lr": 0.0003, "grad_norm": 0.14987953626946238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27485684795, "tokens_per_second": 149245.0930823913, "data_wait_s": 7.321999873965979e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 187900, "loss": 0.5844988469034433, "lr": 0.0003, "grad_norm": 0.15019941374902543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27500261351, "tokens_per_second": 149877.07558403688, "data_wait_s": 1.1980999261140823e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 188000, "loss": 0.6003498584032059, "lr": 0.0003, "grad_norm": 0.15873858486695672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27514881760, "tokens_per_second": 150405.78853102526, "data_wait_s": 9.98800533125177e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 188100, "loss": 0.6589372772723436, "lr": 0.0003, "grad_norm": 0.15250371718394956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27529558837, "tokens_per_second": 150646.49554816974, "data_wait_s": 9.374998626299202e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7dd133210441d33ed680bcdca34cf534ed5191ee691cce94ce99478cd88f06d4
-size 703803
+oid sha256:24b80b49a48da761ea7c382e5dde720e679d73e04560edb56f804ef254e6aee2
+size 705921