Gaston895 commited on
Commit
d097231
Β·
verified Β·
1 Parent(s): e114743

Update README with correct repository structure and enhanced documentation

Browse files
Files changed (1) hide show
  1. README.md +135 -0
README.md CHANGED
@@ -1,3 +1,138 @@
1
  ---
2
  license: apache-2.0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: apache-2.0
3
+ base_model: allura-forge/Llama-3.3-8B-Instruct
4
+ datasets:
5
+ - TeichAI/claude-4.5-opus-high-reasoning-250x
6
+ language:
7
+ - en
8
+ tags:
9
+ - thinking
10
+ - reasoning
11
+ - instruct
12
+ - economics
13
+ - finance
14
+ - analysis
15
+ - llama3.3
16
+ - unsloth
17
+ - finetune
18
+ - bfloat16
19
+ - 128k context
20
+ pipeline_tag: text-generation
21
+ library_name: transformers
22
  ---
23
+
24
+ # AEGIS Conduct - Economic Analysis Model
25
+
26
+ <img src="econ/matrix-neo-reloaded-fight.gif" style="float:right; width:300px; height:300px; padding:10px;">
27
+
28
+ ## Model Overview
29
+
30
+ This repository contains the Llama 3.3 8B Instruct model with thinking capabilities, fine-tuned for economic and financial analysis using Claude 4.5-Opus High Reasoning dataset.
31
+
32
+ **Key Features:**
33
+ - **Thinking Mode**: Automatic activation for complex reasoning
34
+ - **Economic Focus**: Specialized for financial analysis and market insights
35
+ - **128k Context**: Extended context window for comprehensive analysis
36
+ - **Optimized**: Fine-tuned with Unsloth for efficient inference
37
+
38
+ ## Model Details
39
+
40
+ - **Base Model**: allura-forge/Llama-3.3-8B-Instruct
41
+ - **Fine-tuning Dataset**: TeichAI/claude-4.5-opus-high-reasoning-250x
42
+ - **Context Length**: 128k tokens
43
+ - **Training Method**: Unsloth (3 epochs)
44
+ - **Format**: SafeTensors
45
+ - **Precision**: bfloat16
46
+
47
+ ## Repository Structure
48
+
49
+ The model files are organized in the econ/ directory:
50
+
51
+ ```
52
+ econ/
53
+ β”œβ”€β”€ config.json # Model configuration
54
+ β”œβ”€β”€ generation_config.json # Generation parameters
55
+ β”œβ”€β”€ tokenizer.json # Tokenizer vocabulary
56
+ β”œβ”€β”€ tokenizer_config.json # Tokenizer configuration
57
+ β”œβ”€β”€ special_tokens_map.json # Special tokens mapping
58
+ β”œβ”€β”€ chat_template.jinja # Chat template
59
+ β”œβ”€β”€ model.safetensors.index.json # Model index
60
+ β”œβ”€β”€ model-00001-of-00004.safetensors # Model weights (part 1)
61
+ β”œβ”€β”€ model-00002-of-00004.safetensors # Model weights (part 2)
62
+ β”œβ”€β”€ model-00003-of-00004.safetensors # Model weights (part 3)
63
+ β”œβ”€β”€ model-00004-of-00004.safetensors # Model weights (part 4)
64
+ β”œβ”€β”€ reco.py # Model utilities
65
+ └── matrix-neo-reloaded-fight.gif # Visual asset
66
+ ```
67
+
68
+ ## Usage
69
+
70
+ ### Quick Start with Transformers
71
+
72
+ ```python
73
+ from transformers import AutoTokenizer, AutoModelForCausalLM
74
+
75
+ # Load model and tokenizer from econ subdirectory
76
+ tokenizer = AutoTokenizer.from_pretrained("Gaston895/aegisconduct", subfolder="econ")
77
+ model = AutoModelForCausalLM.from_pretrained("Gaston895/aegisconduct", subfolder="econ")
78
+
79
+ # Generate response
80
+ inputs = tokenizer("Analyze the economic impact of inflation on consumer spending:", return_tensors="pt")
81
+ outputs = model.generate(**inputs, max_length=512, temperature=0.7)
82
+ response = tokenizer.decode(outputs[0], skip_special_tokens=True)
83
+ print(response)
84
+ ```
85
+
86
+ ### Thinking Mode Activation
87
+
88
+ The model automatically activates thinking mode for complex reasoning. Use prompts like:
89
+
90
+ ```python
91
+ # These prompts will trigger thinking mode
92
+ prompts = [
93
+ "Think deeply: Analyze the economic implications of rising interest rates",
94
+ "Explain the financial impact of supply chain disruptions",
95
+ "Think through: What are the long-term effects of quantitative easing?"
96
+ ]
97
+ ```
98
+
99
+ ### Recommended Settings
100
+
101
+ - **Temperature**: 0.7
102
+ - **Repetition Penalty**: 1.05
103
+ - **Top-p**: 0.95
104
+ - **Min-p**: 0.05
105
+ - **Top-k**: 40
106
+ - **Context Window**: 4k minimum, 8k+ recommended
107
+
108
+ ## Capabilities
109
+
110
+ This model excels at:
111
+
112
+ - **Economic Analysis**: Market trends, policy impacts, forecasting
113
+ - **Financial Planning**: Investment strategies, risk assessment
114
+ - **Data Interpretation**: Economic indicators, statistical analysis
115
+ - **Policy Analysis**: Regulatory impacts, fiscal policy effects
116
+ - **Global Economics**: International trade, currency analysis
117
+ - **Research**: Academic-level economic reasoning and explanation
118
+
119
+ ## Example Outputs
120
+
121
+ The model provides detailed, step-by-step reasoning for complex economic questions, often showing its "thinking" process before delivering final answers.
122
+
123
+ ## Technical Notes
124
+
125
+ - Optimized for inference with various backends (transformers, llama.cpp, etc.)
126
+ - Supports both instruct and thinking modes
127
+ - No system prompt required (thinking tags self-generate)
128
+ - Compatible with quantization (Q4KS, IQ3_M recommended minimum)
129
+
130
+ ## License
131
+
132
+ Apache 2.0 (inherited from base model)
133
+
134
+ ## Credits
135
+
136
+ - **Base Model**: [allura-forge/Llama-3.3-8B-Instruct](https://huggingface.co/allura-forge/Llama-3.3-8B-Instruct)
137
+ - **Dataset**: [TeichAI/claude-4.5-opus-high-reasoning-250x](https://huggingface.co/datasets/TeichAI/claude-4.5-opus-high-reasoning-250x)
138
+ - **Training Framework**: [Unsloth](https://github.com/unslothai/unsloth)