Spestly commited on
Commit
bcf3d95
·
verified ·
1 Parent(s): 4fce245

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +19 -0
README.md CHANGED
@@ -57,6 +57,25 @@ library_name: transformers
57
  - **Reasoning Improvement:** In this version of Atlas, I have enhanced it's reasoning via synthetic data from models such as Gemini-2.0 Flash Thinking so that it can improve on reasoning.
58
  ---
59
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60
  ### **Intended Use Cases**
61
  Atlas Pro works best for:
62
  - **Technical Professionals:** Helping developers, engineers, and scientists solve complex problems.
 
57
  - **Reasoning Improvement:** In this version of Atlas, I have enhanced it's reasoning via synthetic data from models such as Gemini-2.0 Flash Thinking so that it can improve on reasoning.
58
  ---
59
 
60
+ # **Evaluation**
61
+ Below are the evaluations of the Atlas-Pro models and Deepseek's R1 Qwen Distills (The model that started the whole Atlas family):
62
+
63
+
64
+ | **Metric** | **Spestly Atlas Pro (7B)** | **Spestly Atlas Pro (1.5B)** | DeepSeek-R1-Distill-Qwen (7B) | DeepSeek-R1-Distill-Qwen (1.5B) |
65
+ |-------------------------|---------------------------|------------------------------|-----------------------------------|-------------------------------------|
66
+ | **Average** | **22.65%** | 12.93% | 11.73% | 7.53% |
67
+ | **IFEval** | 31.54% | 24.30% | **40.38%** | 34.63% |
68
+ | **BBH** | **25.27%** | 9.08% | 7.88% | 4.73% |
69
+ | **MATH** | **38.90%** | 25.83% | 0.00% | 0.00% |
70
+ | **GPQA** | **11.63%** | 6.26% | 3.91% | 2.97% |
71
+ | **MUSR** | **6.65%** | 1.86% | 3.55% | 2.08% |
72
+ | **MMLU-Pro** | **21.89%** | 10.28% | 14.68% | 0.78% |
73
+ | **Carbon Emissions (kg)** | 0.69 kg | **0.59 kg** | 0.68 kg | 0.62 kg |
74
+
75
+
76
+
77
+
78
+ ---
79
  ### **Intended Use Cases**
80
  Atlas Pro works best for:
81
  - **Technical Professionals:** Helping developers, engineers, and scientists solve complex problems.