Hugging Face's logo

Charley890
/
AdaptiveMath

Text Generation
Transformers
Safetensors
English
mixtral
mathematics
reasoning
algebra
arithmetic
geometry
calculus
statistics
number-theory
ai
adaption-labs
autoscientist
lora
education
stem
conversational
text-generation-inference
Model card Files
xet
 Community
Charley890 commited on
Commit
19d1bbe
·
verified ·
1 Parent(s): 5f25fbc

Update README.md

Browse files

https://cdn-uploads.huggingface.co/production/uploads/680b3b2640748ca0081daae4/-aHNulF2KnHpaJLDmOk_D.mp4

Files changed (1) hide show
  1. README.md +74 -110
README.md CHANGED
@@ -1,182 +1,146 @@
1
  ---
2
  license: apache-2.0
3
- base_model: mistralai/Mixtral-8x7B
4
  library_name: transformers
5
  pipeline_tag: text-generation
 
6
  tags:
7
  - mathematics
8
  - reasoning
9
- - chain-of-thought
10
  - algebra
11
  - arithmetic
12
  - geometry
13
  - calculus
14
  - statistics
 
15
  - ai
16
  - adaption-labs
17
  - autoscientist
18
  - lora
19
  - education
20
  - stem
 
21
  language:
22
  - en
23
 
24
 
 
25
  metrics:
26
- - Win Rate
 
27
  - Math Win Rate
28
  - Percentile
29
- - Grade
30
- ---
31
-
32
- # 🧮 Adaptive Math – Mixtral-8x7B
33
-
34
- <div align="center">
35
-
36
- # 📐 Adaptive Mathematics Reasoning Model
37
-
38
- **Fine-tuned with Adaption Labs AutoScientist**
39
-
40
- Base Model: **Mixtral-8x7B**
41
-
42
- Specialized for mathematical reasoning, symbolic computation, algebra, arithmetic, geometry, statistics and structured problem solving.
43
-
44
- </div>
45
 
46
  ---
47
 
48
- # 📚 Model Overview
49
-
50
- Adaptive Math is a mathematics-focused reasoning model fine-tuned using **Adaption Labs AutoScientist**.
51
-
52
- The model is designed to improve mathematical reasoning ability across numerous mathematical tasks while maintaining strong logical consistency and step-by-step problem solving.
53
 
54
- Rather than only predicting answers, the model learns structured reasoning patterns that improve mathematical accuracy on numerical, symbolic and logical problems.
55
 
56
  ---
57
 
58
- # ✨ Key Features
59
 
60
- ✅ Mathematical reasoning
61
 
62
- ✅ Step-by-step solutions
63
 
64
- ✅ Arithmetic
65
 
66
- ✅ Algebra
67
 
68
- ✅ Geometry
69
 
70
- ✅ Statistics
 
 
 
 
 
 
 
 
 
71
 
72
- ✅ Number Theory
73
 
74
- ✅ Logical Reasoning
75
 
76
- ✅ Symbolic Mathematics
77
 
78
- ✅ Educational Applications
79
 
80
- ✅ STEM Learning
81
 
82
- ✅ AI-assisted Mathematics
 
 
 
 
 
 
 
83
 
84
  ---
85
 
86
- # 🎯 Intended Uses
87
 
88
- This model is suitable for:
89
-
90
- - Mathematics tutoring
91
- - Educational assistants
92
- - Homework support
93
- - AI research
94
- - Mathematical reasoning
95
- - Numerical analysis
96
- - Scientific assistants
97
- - Coding assistants requiring mathematics
98
- - Data science education
99
- - Engineering calculations
100
 
101
  ---
102
 
103
- # ⚙️ Base Model
104
 
105
  | Property | Value |
106
- |-----------|--------|
107
- | Base Model | Mixtral-8x7B |
108
- | Fine-tuning | Adaption Labs AutoScientist |
109
- | Task | Mathematical Reasoning |
110
- | Framework | Transformers |
111
  | Language | English |
112
 
113
  ---
114
 
115
- # 📈 Training Performance
116
-
117
- The model was optimized using **Adaption Labs AutoScientist**.
118
-
119
- ## Overall Performance
120
-
121
- | Metric | Before | After |
122
- |--------|--------:|-------:|
123
- | Grade | E | **A** |
124
- | Percentile | 0.0 | **57.7** |
125
- | Quality Score | 5.0 | **9.1** |
126
-
127
- ---
128
-
129
- ## AutoScientist Results
130
-
131
- | Metric | Base Model | Adaptive Model |
132
- |---------|-----------:|---------------:|
133
- | Win Rate | 19 | **81** |
134
- | Math Win Rate | 38 | **63** |
135
 
136
- ---
137
-
138
- # 📊 Training Summary
139
 
140
- | Component | Status |
141
- |-----------|--------|
142
- | Fine-tuning | ✅ Completed |
143
- | AutoScientist Optimization | ✅ Completed |
144
- | Mathematical Evaluation | ✅ Completed |
145
- | Grade Improvement | E → A |
146
- | Relative Quality Improvement | 82% |
147
 
148
  ---
149
 
150
- # 📉 Training Metrics
151
-
152
- The optimization process monitored several training signals including:
153
 
154
- - Training Loss
155
- - Validation Loss
156
- - Learning Rate
157
- - Gradient Norm
158
- - Win Rate
159
- - Math Win Rate
160
- - Percentile
161
- - Overall Grade
162
 
163
- These metrics demonstrate significant improvements after AutoScientist optimization.
164
 
165
- ---
166
 
167
- # 🚀 Why this model?
168
 
169
- Compared with the original base model, Adaptive Math demonstrates:
170
 
171
- - Higher mathematical accuracy
172
- - Better reasoning consistency
173
- - Improved symbolic manipulation
174
- - Better numerical understanding
175
- - Improved mathematical generalization
176
- - Stronger educational usefulness
177
 
 
178
 
 
 
179
 
180
 
181
- ```
182
- Adaptive Math.mp4
 
1
  ---
2
  license: apache-2.0
3
+ base_model: mistralai/Mixtral-8x7B-Instruct-v0.1
4
  library_name: transformers
5
  pipeline_tag: text-generation
6
+
7
  tags:
8
  - mathematics
9
  - reasoning
 
10
  - algebra
11
  - arithmetic
12
  - geometry
13
  - calculus
14
  - statistics
15
+ - number-theory
16
  - ai
17
  - adaption-labs
18
  - autoscientist
19
  - lora
20
  - education
21
  - stem
22
+
23
  language:
24
  - en
25
 
26
 
27
+
28
  metrics:
29
+ - Grade
30
+ - Overall Win Rate
31
  - Math Win Rate
32
  - Percentile
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33
 
34
  ---
35
 
36
+ # 🧮 Adaptive Math — Mixtral-8x7B
 
 
 
 
37
 
38
+ > A PEFT (LoRA) fine-tuned **Mixtral-8x7B-Instruct-v0.1** model specialized for mathematical reasoning, arithmetic, algebra, geometry, statistics and logical problem solving using the **Adaption Labs AutoScientist** workflow.
39
 
40
  ---
41
 
42
+ # Overview
43
 
44
+ Adaptive Math is a domain-adapted language model developed for mathematical reasoning and instruction following.
45
 
46
+ The model was fine-tuned using the **Adaption Labs AutoScientist** workflow on a mathematics reasoning dataset containing thousands of mathematical instruction-completion pairs.
47
 
48
+ Unlike a general language model, this model focuses on structured mathematical reasoning, symbolic manipulation and logical step-by-step solutions.
49
 
50
+ ---
51
 
52
+ # Quick Facts
53
 
54
+ | Item | Value |
55
+ |------|-------|
56
+ | Domain | Mathematics |
57
+ | Task | Mathematical Reasoning |
58
+ | Base Model | Mixtral-8x7B-Instruct-v0.1 |
59
+ | Fine-tuning Method | PEFT (LoRA) |
60
+ | Training Framework | Adaption Labs AutoScientist |
61
+ | Dataset | adaption-single-integer-samples |
62
+ | Language | English |
63
+ | License | Apache-2.0 |
64
 
65
+ ---
66
 
67
+ # Training Dataset
68
 
69
+ Dataset Repository
70
 
71
+ https://huggingface.co/datasets/Charley890/adaption-single-integer-samples
72
 
73
+ The training dataset contains mathematical reasoning examples covering:
74
 
75
+ - Arithmetic
76
+ - Algebra
77
+ - Number Theory
78
+ - Geometry
79
+ - Statistics
80
+ - Logical Reasoning
81
+ - Symbolic Mathematics
82
+ - Mathematical Word Problems
83
 
84
  ---
85
 
86
+ # Key Features
87
 
88
+ - Mathematical Question Answering
89
+ - Step-by-Step Reasoning
90
+ - Arithmetic Computation
91
+ - Algebraic Manipulation
92
+ - Geometry Reasoning
93
+ - Number Theory
94
+ - Statistics
95
+ - Symbolic Reasoning
96
+ - STEM Education
97
+ - AI Tutoring
 
 
98
 
99
  ---
100
 
101
+ # Base Model
102
 
103
  | Property | Value |
104
+ |-----------|-------|
105
+ | Base Model | Mixtral-8x7B-Instruct-v0.1 |
106
+ | Fine-tuning | PEFT (LoRA) |
107
+ | Framework | Adaption Labs AutoScientist |
 
108
  | Language | English |
109
 
110
  ---
111
 
112
+ # Evaluation Results
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
113
 
114
+ Evaluation performed using **Adaption Labs AutoScientist**
 
 
115
 
116
+ | Metric | Base Model | Adapted Model | Improvement |
117
+ |--------|-----------:|--------------:|------------:|
118
+ | Grade | E | A | Excellent |
119
+ | Overall Win Rate | 19 | 81 | +62 |
120
+ | Math Win Rate | 38 | 63 | +25 |
121
+ | Percentile | 0.0 | 57.7 | +57.7 |
122
+ | Relative Improvement | — | — | 82% |
123
 
124
  ---
125
 
126
+ # Performance Summary
 
 
127
 
128
+ ```
129
+ ✓ Training Completed Successfully
 
 
 
 
 
 
130
 
131
+ ✓ Stable Optimization
132
 
133
+ ✓ Mathematical Reasoning Improved
134
 
135
+ ✓ Grade Improved from E → A
136
 
137
+ ✓ Overall Win Rate Improved from 19 → 81
138
 
139
+ ✓ Math Win Rate Improved from 38 → 63
 
 
 
 
 
140
 
141
+ ✓ Percentile Improved from 0.0 → 57.7
142
 
143
+ ✓ Relative Quality Improvement: 82%
144
+ ```
145
 
146