Daemontatox
commited on
Commit
•
599585d
1
Parent(s):
de049a1
Update README.md
Browse files
README.md
CHANGED
@@ -1,22 +1,130 @@
|
|
1 |
---
|
2 |
-
base_model:
|
|
|
3 |
tags:
|
4 |
- text-generation-inference
|
5 |
- transformers
|
6 |
- unsloth
|
7 |
-
-
|
8 |
- trl
|
|
|
|
|
9 |
license: apache-2.0
|
10 |
language:
|
11 |
- en
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
---
|
13 |
|
14 |
-
|
15 |
|
16 |
-
|
17 |
-
- **License:** apache-2.0
|
18 |
-
- **Finetuned from model :** unsloth/qwq-32b-preview-bnb-4bit
|
19 |
|
20 |
-
|
|
|
|
|
21 |
|
22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
base_model:
|
3 |
+
- Meta/LLama3.1-8B-Instruct
|
4 |
tags:
|
5 |
- text-generation-inference
|
6 |
- transformers
|
7 |
- unsloth
|
8 |
+
- Llama3
|
9 |
- trl
|
10 |
+
- COT
|
11 |
+
- Reasoning
|
12 |
license: apache-2.0
|
13 |
language:
|
14 |
- en
|
15 |
+
datasets:
|
16 |
+
- Daemontatox/LongCOT-Reason
|
17 |
+
metrics:
|
18 |
+
- accuracy
|
19 |
+
- character
|
20 |
+
- competition_math
|
21 |
+
- code_eval
|
22 |
+
library_name: transformers
|
23 |
+
new_version: Daemontatox/AetherLlama
|
24 |
+
pipeline_tag: text-generation
|
25 |
---
|
26 |
|
27 |
+
![image](./image.webp)
|
28 |
|
29 |
+
# AetherLlama
|
|
|
|
|
30 |
|
31 |
+
- **Developed by:** Daemontatox
|
32 |
+
- **License:** Apache 2.0
|
33 |
+
- **Finetuned Using:** [Unsloth](https://github.com/unslothai/unsloth), Hugging Face Transformers, and TRL Library
|
34 |
|
35 |
+
## Model Overview
|
36 |
+
|
37 |
+
The ** AetherLlama Model** is an advanced AI system optimized for logical reasoning, multi-step problem-solving, and decision-making tasks. Designed with efficiency and accuracy in mind, it employs a structured system prompt to ensure high-quality answers through a transparent and iterative thought process.
|
38 |
+
|
39 |
+
### System Prompt and Workflow
|
40 |
+
|
41 |
+
This model operates using an innovative reasoning framework structured around the following steps:
|
42 |
+
|
43 |
+
1. **Initial Thought:**
|
44 |
+
The model uses `<Thinking>` tags to reason step-by-step and craft its best possible response.
|
45 |
+
Example:
|
46 |
+
|
47 |
+
2. **Self-Critique:**
|
48 |
+
It evaluates its initial response within `<Critique>` tags, focusing on:
|
49 |
+
- **Accuracy:** Is it factually correct and verifiable?
|
50 |
+
- **Clarity:** Is it clear and free of ambiguity?
|
51 |
+
- **Completeness:** Does it fully address the request?
|
52 |
+
- **Improvement:** What can be enhanced?
|
53 |
+
Example:
|
54 |
+
|
55 |
+
3. **Revision:**
|
56 |
+
Based on the critique, the model refines its response within `<Revising>` tags.
|
57 |
+
Example:
|
58 |
+
|
59 |
+
4. **Final Response:**
|
60 |
+
The revised response is presented clearly within `<Final>` tags.
|
61 |
+
Example:
|
62 |
+
|
63 |
+
5. **Tag Innovation:**
|
64 |
+
When needed, the model creates and defines new tags for better structuring or clarity, ensuring consistent usage.
|
65 |
+
Example:
|
66 |
+
|
67 |
+
### Key Features
|
68 |
+
- **Structured Reasoning:** Transparent, multi-step approach for generating and refining answers.
|
69 |
+
- **Self-Improvement:** Built-in critique and revision ensure continuous response enhancement.
|
70 |
+
- **Clarity and Adaptability:** Tagging system provides organized, adaptable responses tailored to user needs.
|
71 |
+
- **Creative Flexibility:** Supports dynamic problem-solving with the ability to introduce new tags and concepts.
|
72 |
+
|
73 |
+
---
|
74 |
+
|
75 |
+
## Use Cases
|
76 |
+
|
77 |
+
The model is designed for various domains, including:
|
78 |
+
1. **Research and Analysis:** Extracting insights and providing structured explanations.
|
79 |
+
2. **Education:** Assisting with tutoring by breaking down complex problems step-by-step.
|
80 |
+
3. **Problem-Solving:** Offering logical and actionable solutions for multi-step challenges.
|
81 |
+
4. **Content Generation:** Producing clear, well-organized creative or professional content.
|
82 |
+
|
83 |
+
---
|
84 |
+
|
85 |
+
## Training Details
|
86 |
+
|
87 |
+
- **Frameworks:**
|
88 |
+
- [Unsloth](https://github.com/unslothai/unsloth) for accelerated training.
|
89 |
+
- Hugging Face Transformers and the TRL library for reinforcement learning with human feedback (RLHF).
|
90 |
+
|
91 |
+
- **Dataset:** Finetuned on diverse reasoning-focused tasks, including logical puzzles, mathematical problems, and commonsense reasoning scenarios.
|
92 |
+
|
93 |
+
- **Hardware Efficiency:**
|
94 |
+
- Trained with bnb-4bit precision for reduced memory usage.
|
95 |
+
- Optimized training pipeline achieving 2x faster development cycles.
|
96 |
+
|
97 |
+
---
|
98 |
+
|
99 |
+
## Performance Metrics
|
100 |
+
|
101 |
+
The model excels in reasoning benchmarks:
|
102 |
+
- **ARC (AI2 Reasoning Challenge):** High accuracy in logical and commonsense tasks.
|
103 |
+
- **GSM8K (Math Reasoning):** Superior results in multi-step problem-solving.
|
104 |
+
- **CommonsenseQA:** Strong comprehension of everyday reasoning tasks.
|
105 |
+
|
106 |
+
---
|
107 |
+
|
108 |
+
## Ethical Considerations
|
109 |
+
|
110 |
+
- **Transparency:** Responses are structured for verifiability through tagging.
|
111 |
+
- **Bias Mitigation:** Includes self-critique to minimize biases and ensure fairness.
|
112 |
+
- **Safe Deployment:** Users are encouraged to evaluate outputs to prevent harm or misinformation.
|
113 |
+
|
114 |
+
---
|
115 |
+
|
116 |
+
## License
|
117 |
+
|
118 |
+
This model is distributed under the Apache 2.0 license, allowing users to use, modify, and share it in compliance with the license terms.
|
119 |
+
|
120 |
+
---
|
121 |
+
|
122 |
+
## Acknowledgments
|
123 |
+
|
124 |
+
Special thanks to:
|
125 |
+
- [Unsloth](https://github.com/unslothai/unsloth) for accelerated training workflows.
|
126 |
+
- Hugging Face for their powerful tools and libraries.
|
127 |
+
|
128 |
+
---
|
129 |
+
|
130 |
+
Experience the **AetherLlama l**, leveraging its structured reasoning and self-improvement capabilities for any task requiring advanced AI reasoning.
|