alex/club-3090
1
0
Fork 0
Code Issues Pull Requests Actions 38 Packages Projects Releases Wiki Activity

chore(gitignore): allow results/lucebox-*/ — evidence for BENCHMARKS lucebox row

Browse Source 
Mirrors the existing negations for grammar-ab-* / grammar-full-* / v0.20-migration.
The lucebox-dual-gpu-20260504-142832/ run from cb089e1 backs the published
BENCHMARKS row; tracking it here keeps the evidence reproducible.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
noonghunna
2026-05-04 14:33:49 +00:00
parent cb089e1e36
commit 030f780f24
3 changed files with 60 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
results/lucebox-dual-gpu-20260504-142832/phase-a-same-card.log Normal file
View File

@@ -0,0 +1,35 @@
[transformers] PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
[bench] target = /mnt/models/gguf/qwen3.5-27b/Qwen3.5-27B-Q4_K_M.gguf
[bench] draft = /mnt/models/huggingface/qwen3.5-27b-dflash/model.safetensors
[bench] bin = /opt/ai/lucebox-hub/dflash/build/test_dflash
[bench] tmp = /tmp/dflash_bench
[bench] tokenizing prompts via HF…
[00] has_close_elements 125 tokens
[01] separate_paren_groups 135 tokens
[02] truncate_number 86 tokens
[03] below_zero 128 tokens
[04] mean_absolute_deviation 120 tokens
[05] intersperse 104 tokens
[06] parse_nested_parens 139 tokens
[07] filter_by_substring 95 tokens
[08] sum_product 131 tokens
[09] rolling_max 120 tokens
[bench] mode=fast n_gen=128
prompt steps AL pct% tok/s
--------------------------------------------------------------
has_close_elements 28 4.57 28.8 54.36
separate_paren_groups 21 6.10 39.0 71.84
truncate_number 22 5.82 38.4 69.30
below_zero 23 5.57 34.8 66.00
mean_absolute_deviation 26 4.92 31.2 57.97
intersperse 29 4.41 28.4 52.72
parse_nested_parens 13 9.85 63.5 108.74
filter_by_substring 23 5.57 35.6 65.75
sum_product 15 8.53 59.2 96.55
rolling_max 15 8.53 54.2 96.44
--------------------------------------------------------------
MEAN 6.39 41.3 73.97
commit/step range: 4.41 - 9.85
tok/s range: 52.7 - 108.7

24
results/lucebox-dual-gpu-20260504-142832/phase-b-dual-gpu.log Normal file
View File

@@ -0,0 +1,24 @@
[bench] target = /mnt/models/gguf/qwen3.5-27b/Qwen3.5-27B-Q4_K_M.gguf
[bench] draft = /mnt/models/huggingface/qwen3.5-27b-dflash/model.safetensors
[bench] bin = /opt/ai/lucebox-hub/dflash/build/test_dflash
[bench] tmp = /tmp/dflash_bench
[bench] skipping tokenize (reusing /tmp/dflash_bench)
[bench] mode=fast n_gen=128
prompt steps AL pct% tok/s
--------------------------------------------------------------
has_close_elements 28 4.57 28.8 55.90
separate_paren_groups 21 6.10 39.0 72.55
truncate_number 22 5.82 38.4 70.58
below_zero 23 5.57 34.8 67.09
mean_absolute_deviation 26 4.92 31.2 59.88
intersperse 29 4.41 28.4 54.16
parse_nested_parens 13 9.85 63.5 109.98
filter_by_substring 23 5.57 35.6 66.47
sum_product 15 8.53 59.2 97.52
rolling_max 15 8.53 54.2 98.23
--------------------------------------------------------------
MEAN 6.39 41.3 75.24
commit/step range: 4.41 - 9.85
tok/s range: 54.2 - 110.0