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": 74891.76216789891,
-      "required_pages_per_second_upper_estimate": 0.006491428133277603
+      "required_token_positions_per_second": 75437.76506631428,
+      "required_pages_per_second_upper_estimate": 0.006538754280680553
     },
     {
       "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": 9361.470270987364,
-      "required_pages_per_second_upper_estimate": 0.0037344304575504085
+      "required_token_positions_per_second": 9429.720633289286,
+      "required_pages_per_second_upper_estimate": 0.0037616565475064967
     },
     {
       "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": 18722.940541974727,
-      "required_pages_per_second_upper_estimate": 0.004331222781886563
+      "required_token_positions_per_second": 18859.44126657857,
+      "required_pages_per_second_upper_estimate": 0.00436279982219296
     },
     {
       "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": 28084.41081296209,
-      "required_pages_per_second_upper_estimate": 0.0017101932709934718
+      "required_token_positions_per_second": 28289.161899867857,
+      "required_pages_per_second_upper_estimate": 0.001722661537940102
     },
     {
       "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": 28084.41081296209,
-      "required_pages_per_second_upper_estimate": 0.0021620156870067013
+      "required_token_positions_per_second": 28289.161899867857,
+      "required_pages_per_second_upper_estimate": 0.0021777779924639917
     },
     {
       "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": 28084.41081296209,
-      "required_pages_per_second_upper_estimate": 0.34871438982035696
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28289.161899867857,
+      "required_pages_per_second_upper_estimate": 0.3512567130619201
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.2516666526252038,
-    "producer_p95_seconds": 0.3799248219947913,
-    "gpu_update_seconds": 0.9844607452914529,
-    "producer_batches_per_second": 3.9735101554724315,
-    "gpu_batches_per_second": 1.0157845346123442,
-    "producer_headroom": 3.9117647690795465,
+    "producer_batch_seconds": 0.15208074697659413,
+    "producer_p95_seconds": 0.3972021820009104,
+    "gpu_update_seconds": 0.9773354225856068,
+    "producer_batches_per_second": 6.575454289121188,
+    "gpu_batches_per_second": 1.023190172882952,
+    "producer_headroom": 6.426424396350597,
     "sustainable": true,
-    "buffer_seconds": 60.05210546277863,
+    "buffer_seconds": 60.59479620030762,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377141.13390771794,
-    "upload_bytes_per_second": 6015192.468192719,
+    "generation_bytes_per_second": 376757.7644130556,
+    "upload_bytes_per_second": 5947384.905171534,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.9257079403272,
+    "estimated_full_shards_per_hour": 80.84344577115776,
     "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": 0.9512047030002577,
-      "bytes": 7497853,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1582007809993229,
       "bytes": 12016223,
@@ -562,6 +556,12 @@
       "bytes": 16394940,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 0.905758473003516,
+      "bytes": 2394822,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -313,3 +313,4 @@
 {"step": 116300, "loss": 0.6128046456724405, "lr": 0.0003, "grad_norm": 0.15431430011580516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17018720502, "tokens_per_second": 148930.9296571331, "data_wait_s": 2.039800165221095e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 116400, "loss": 0.5653275307267904, "lr": 0.0003, "grad_norm": 0.13909962304253426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17033279015, "tokens_per_second": 133019.79481446205, "data_wait_s": 1.032499858411029e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 116500, "loss": 0.6135224271565676, "lr": 0.0003, "grad_norm": 0.13826183063391081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17047979953, "tokens_per_second": 149529.03783741716, "data_wait_s": 1.65310048032552e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 116600, "loss": 0.6222167238593102, "lr": 0.0003, "grad_norm": 0.14175215385024614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17062709580, "tokens_per_second": 149558.18056459204, "data_wait_s": 1.7295002180617303e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:35b98dc572c8d4b6de55cf2a7e63161391eaf088d04a6bcfef76ed47378bf5a0
-size 672033
+oid sha256:840fa06bcbea020749d217cf7683c7ad22b045e4e82bfb61ea562c103bb56fc5
+size 674151