File size: 1,127 Bytes
d473371
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
# AUP Benchmark Data
# Format: task -> method -> list of (rho, accuracy) pairs
# rho: parallelism (tokens per forward)
# accuracy: model accuracy (0-1 scale)

# Model metadata: type (AR/dLLM), foundation model, link
_meta:
  Qwen2.5-Coder-7B-it:
    type: AR
    foundation: Qwen2.5-Coder-7B-it
    link: https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct
  Dream-Coder-7B:
    type: dLLM
    foundation: Dream-Coder-v0-it-7B
    link: https://github.com/DreamLM/Dream-Coder
  d3LLM-Coder-7B:
    type: dLLM
    foundation: Dream-Coder-v0-it-7B
    link: https://github.com/hao-ai-lab/d3llm

HumanEval:
  Qwen2.5-Coder-7B-it:
  - [1.0, 86.6]
  Dream-Coder-7B:
  - [1.0, 82.9]
  d3LLM-Coder-7B:
  - [1.0, 82.4]
  - [2.88, 79.7]
HumanEval+:
  Qwen2.5-Coder-7B-it:
  - [1.0, 82.3]
  Dream-Coder-7B:
  - [1.0, 76.8]
  d3LLM-Coder-7B:
  - [1.0, 74.4]
  - [2.88, 71.3]
MBPP:
  Qwen2.5-Coder-7B-it:
  - [1.0, 83.5]
  Dream-Coder-7B:
  - [1.0, 79.9]
  d3LLM-Coder-7B:
  - [1.0, 80.10]
  - [2.5, 80.00]
MBPP+:
  Qwen2.5-Coder-7B-it:
  - [1.0, 70.1]
  Dream-Coder-7B:
  - [1.0, 68.8]
  d3LLM-Coder-7B:
  - [1.0, 69.6]
  - [2.5, 69.3]