Dataset Viewer
Auto-converted to Parquet Duplicate
model_id
large_stringclasses
6 values
benchmark
large_stringclasses
4 values
scenario
large_stringclasses
4 values
n_problems
int64
164
1.06k
pass_at_1
float64
0.26
0.91
pass_at_1_ci_low
float64
0.23
0.87
pass_at_1_ci_high
float64
0.28
0.96
pass_at_1_base
float64
0.63
0.94
pass_at_1_plus
float64
0.57
0.91
n_samples_per_problem
int64
1
1
sampling
large_stringclasses
1 value
hub_license
large_stringclasses
3 values
osi_approved
bool
2 classes
harness
large_stringclasses
3 values
serving
large_stringclasses
1 value
dtype
large_stringclasses
1 value
contamination
large_stringclasses
3 values
pass_at_1_clean
float64
0.18
0.3
n_clean
float64
0
535
Qwen/Qwen2.5-Coder-32B-Instruct
cruxeval
cruxeval_i
800
0.6787
0.645
0.7113
null
null
1
greedy(temp=0)
apache-2.0
true
cruxeval(direct,2-shot)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen2.5-Coder-14B-Instruct
cruxeval
cruxeval_i
800
0.63
0.5962
0.6637
null
null
1
greedy(temp=0)
apache-2.0
true
cruxeval(direct,2-shot)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen3-Coder-30B-A3B-Instruct
cruxeval
cruxeval_i
800
0.6138
0.58
0.6475
null
null
1
greedy(temp=0)
apache-2.0
true
cruxeval(direct,2-shot)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen2.5-Coder-7B-Instruct
cruxeval
cruxeval_i
800
0.5487
0.5138
0.5825
null
null
1
greedy(temp=0)
apache-2.0
true
cruxeval(direct,2-shot)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct
cruxeval
cruxeval_i
800
0.4037
0.37
0.4375
null
null
1
greedy(temp=0)
deepseek-license (non-OSI)
false
cruxeval(direct,2-shot)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen3-Coder-30B-A3B-Instruct
cruxeval
cruxeval_o
800
0.6587
0.6262
0.6925
null
null
1
greedy(temp=0)
apache-2.0
true
cruxeval(direct,2-shot)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen2.5-Coder-32B-Instruct
cruxeval
cruxeval_o
800
0.6412
0.6075
0.675
null
null
1
greedy(temp=0)
apache-2.0
true
cruxeval(direct,2-shot)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen2.5-Coder-14B-Instruct
cruxeval
cruxeval_o
800
0.5988
0.5637
0.6338
null
null
1
greedy(temp=0)
apache-2.0
true
cruxeval(direct,2-shot)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen2.5-Coder-7B-Instruct
cruxeval
cruxeval_o
800
0.5225
0.4888
0.5575
null
null
1
greedy(temp=0)
apache-2.0
true
cruxeval(direct,2-shot)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct
cruxeval
cruxeval_o
800
0.4487
0.415
0.4825
null
null
1
greedy(temp=0)
deepseek-license (non-OSI)
false
cruxeval(direct,2-shot)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen3-Coder-30B-A3B-Instruct
humanevalplus
default
164
0.909
0.872
0.9573
0.939
0.909
1
greedy(temp=0)
apache-2.0
true
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen2.5-Coder-32B-Instruct
humanevalplus
default
164
0.866
0.8171
0.9207
0.89
0.866
1
greedy(temp=0)
apache-2.0
true
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen2.5-Coder-14B-Instruct
humanevalplus
default
164
0.823
0.7622
0.878
0.878
0.823
1
greedy(temp=0)
apache-2.0
true
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen2.5-Coder-7B-Instruct
humanevalplus
default
164
0.799
0.7378
0.8598
0.854
0.799
1
greedy(temp=0)
apache-2.0
true
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct
humanevalplus
default
164
0.756
0.7011
0.8232
0.793
0.756
1
greedy(temp=0)
deepseek-license (non-OSI)
false
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
bigcode/starcoder2-15b-instruct-v0.1
humanevalplus
default
164
0.573
0.5061
0.6585
0.634
0.573
1
greedy(temp=0)
bigcode-openrail-m (non-OSI)
false
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen3-Coder-30B-A3B-Instruct
livecodebench
codegeneration
1,055
0.5318
0.5014
0.5621
null
null
1
greedy(temp=0)
apache-2.0
true
livecodebench(vendored official codegen_metrics)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
NO clean window (model released after v6 ends)
null
0
Qwen/Qwen2.5-Coder-32B-Instruct
livecodebench
codegeneration
1,055
0.4199
0.3905
0.4493
null
null
1
greedy(temp=0)
apache-2.0
true
livecodebench(vendored official codegen_metrics)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
date-windowed clean subset reported (contest_date > model cutoff)
0.3047
535
Qwen/Qwen2.5-Coder-14B-Instruct
livecodebench
codegeneration
1,055
0.3659
0.3374
0.3953
null
null
1
greedy(temp=0)
apache-2.0
true
livecodebench(vendored official codegen_metrics)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
date-windowed clean subset reported (contest_date > model cutoff)
0.2262
535
Qwen/Qwen2.5-Coder-7B-Instruct
livecodebench
codegeneration
1,055
0.2891
0.2616
0.3175
null
null
1
greedy(temp=0)
apache-2.0
true
livecodebench(vendored official codegen_metrics)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
date-windowed clean subset reported (contest_date > model cutoff)
0.1757
535
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct
livecodebench
codegeneration
1,055
0.255
0.2294
0.2815
null
null
1
greedy(temp=0)
deepseek-license (non-OSI)
false
livecodebench(vendored official codegen_metrics)
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
NO clean window (model released after v6 ends)
null
0
Qwen/Qwen2.5-Coder-32B-Instruct
mbppplus
default
378
0.767
0.7302
0.8122
0.886
0.767
1
greedy(temp=0)
apache-2.0
true
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen3-Coder-30B-A3B-Instruct
mbppplus
default
378
0.765
0.7222
0.8069
0.889
0.765
1
greedy(temp=0)
apache-2.0
true
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen2.5-Coder-14B-Instruct
mbppplus
default
378
0.743
0.7063
0.7937
0.862
0.743
1
greedy(temp=0)
apache-2.0
true
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct
mbppplus
default
378
0.693
0.6508
0.7434
0.817
0.693
1
greedy(temp=0)
deepseek-license (non-OSI)
false
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
Qwen/Qwen2.5-Coder-7B-Instruct
mbppplus
default
378
0.659
0.6217
0.7169
0.765
0.659
1
greedy(temp=0)
apache-2.0
true
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null
bigcode/starcoder2-15b-instruct-v0.1
mbppplus
default
378
0.651
0.6058
0.7011
0.77
0.651
1
greedy(temp=0)
bigcode-openrail-m (non-OSI)
false
evalplus
vllm-0.23.0 (apptainer, sm_120 RTX Pro 6000)
bfloat16
regression-only (likely in pretrain)
null
null

benchForge Leaderboard (v1)

Reproducible code-LLM evaluation forged on idle TU Delft DAIC RTX Pro 6000 (Blackwell, sm_120) GPUs. vLLM 0.23 (apptainer), bf16, greedy temp=0, pass@1 with bootstrap-over-problems 95% CI. HumanEval+/MBPP+ via EvalPlus (pass_at_1=plus set); CRUXEval-I/-O = input/output prediction, Direct (no-CoT), 2-shot, execution-scored (comparable to the paper's Direct column).

Honest caveat: HumanEval+/MBPP+/CRUXEval are likely in pretraining corpora -> comparison/regression numbers, not contamination-free. Method+code: github.com/D4vidHuang/benchForge.

model_id benchmark scenario pass_at_1 pass_at_1_ci_low pass_at_1_ci_high n_problems
Qwen/Qwen2.5-Coder-32B-Instruct cruxeval cruxeval_i 0.6787 0.645 0.7113 800
Qwen/Qwen2.5-Coder-14B-Instruct cruxeval cruxeval_i 0.63 0.5962 0.6637 800
Qwen/Qwen3-Coder-30B-A3B-Instruct cruxeval cruxeval_i 0.6138 0.58 0.6475 800
Qwen/Qwen2.5-Coder-7B-Instruct cruxeval cruxeval_i 0.5487 0.5138 0.5825 800
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct cruxeval cruxeval_i 0.4037 0.37 0.4375 800
Qwen/Qwen3-Coder-30B-A3B-Instruct cruxeval cruxeval_o 0.6587 0.6262 0.6925 800
Qwen/Qwen2.5-Coder-32B-Instruct cruxeval cruxeval_o 0.6412 0.6075 0.675 800
Qwen/Qwen2.5-Coder-14B-Instruct cruxeval cruxeval_o 0.5988 0.5637 0.6338 800
Qwen/Qwen2.5-Coder-7B-Instruct cruxeval cruxeval_o 0.5225 0.4888 0.5575 800
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct cruxeval cruxeval_o 0.4487 0.415 0.4825 800
Qwen/Qwen3-Coder-30B-A3B-Instruct humanevalplus default 0.909 0.872 0.9573 164
Qwen/Qwen2.5-Coder-32B-Instruct humanevalplus default 0.866 0.8171 0.9207 164
Qwen/Qwen2.5-Coder-14B-Instruct humanevalplus default 0.823 0.7622 0.878 164
Qwen/Qwen2.5-Coder-7B-Instruct humanevalplus default 0.799 0.7378 0.8598 164
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct humanevalplus default 0.756 0.7011 0.8232 164
bigcode/starcoder2-15b-instruct-v0.1 humanevalplus default 0.573 0.5061 0.6585 164
Qwen/Qwen3-Coder-30B-A3B-Instruct livecodebench codegeneration 0.5318 0.5014 0.5621 1055
Qwen/Qwen2.5-Coder-32B-Instruct livecodebench codegeneration 0.4199 0.3905 0.4493 1055
Qwen/Qwen2.5-Coder-14B-Instruct livecodebench codegeneration 0.3659 0.3374 0.3953 1055
Qwen/Qwen2.5-Coder-7B-Instruct livecodebench codegeneration 0.2891 0.2616 0.3175 1055
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct livecodebench codegeneration 0.255 0.2294 0.2815 1055
Qwen/Qwen2.5-Coder-32B-Instruct mbppplus default 0.767 0.7302 0.8122 378
Qwen/Qwen3-Coder-30B-A3B-Instruct mbppplus default 0.765 0.7222 0.8069 378
Qwen/Qwen2.5-Coder-14B-Instruct mbppplus default 0.743 0.7063 0.7937 378
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct mbppplus default 0.693 0.6508 0.7434 378
Qwen/Qwen2.5-Coder-7B-Instruct mbppplus default 0.659 0.6217 0.7169 378
bigcode/starcoder2-15b-instruct-v0.1 mbppplus default 0.651 0.6058 0.7011 378
Downloads last month
57

Space using D4vidHuang/benchforge-leaderboard 1