Update README.md
Browse files
README.md
CHANGED
@@ -6,17 +6,141 @@ tags:
|
|
6 |
- unsloth
|
7 |
- gemma3n
|
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/gemma-3n-e2b-it-unsloth-bnb-4bit
|
19 |
|
20 |
This gemma3n model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
21 |
|
22 |
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
- unsloth
|
7 |
- gemma3n
|
8 |
- trl
|
|
|
9 |
language:
|
10 |
- en
|
11 |
---
|
12 |
|
13 |
+
# MindSlate: Fine-tuned Gemma-3B for Personal Knowledge Management
|
14 |
|
15 |
+
|
16 |
+
## Model Description
|
17 |
+
|
18 |
+
**MindSlate** is a fine-tuned version of Google's Gemma-3B model, optimized for personal knowledge management tasks including flashcard generation, reminder processing, content summarization, and task management. The model was trained using Unsloth's efficient fine-tuning techniques for 2x faster training.
|
19 |
+
|
20 |
+
- **Architecture**: Gemma-3B with LoRA adapters
|
21 |
+
- **Model type**: Causal Language Model
|
22 |
+
- **Fine-tuning method**: 4-bit QLoRA
|
23 |
+
- **Languages**: English
|
24 |
+
- **License**: Apache 2.0
|
25 |
+
|
26 |
+
- **Developed by:** [Srinivas Nampalli](https://www.linkedin.com/in/srinivas-nampalli/)
|
27 |
- **License:** apache-2.0
|
28 |
- **Finetuned from model :** unsloth/gemma-3n-e2b-it-unsloth-bnb-4bit
|
29 |
|
30 |
This gemma3n model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
31 |
|
32 |
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
33 |
+
|
34 |
+
## Model Sources
|
35 |
+
|
36 |
+
- **Repository**: [https://github.com/Srinivasmec26/MindSlate](https://github.com/Srinivasmec26/MindSlate)
|
37 |
+
- **Base Model**: [unsloth/gemma-3b-E2B-it-unsloth-bnb-4bit](https://huggingface.co/unsloth/gemma-3b-E2B-it-unsloth-bnb-4bit)
|
38 |
+
|
39 |
+
## Uses
|
40 |
+
|
41 |
+
### Direct Use
|
42 |
+
|
43 |
+
MindSlate is designed for:
|
44 |
+
- Automatic flashcard generation from study materials
|
45 |
+
- Intelligent reminder creation
|
46 |
+
- Content summarization
|
47 |
+
- Task extraction and organization
|
48 |
+
- Personal knowledge base management
|
49 |
+
|
50 |
+
### Downstream Use
|
51 |
+
|
52 |
+
Can be integrated into:
|
53 |
+
- Educational platforms
|
54 |
+
- Productivity apps
|
55 |
+
- Note-taking applications
|
56 |
+
- Personal AI assistants
|
57 |
+
|
58 |
+
### Out-of-Scope Use
|
59 |
+
|
60 |
+
Not suitable for:
|
61 |
+
- Medical or legal advice
|
62 |
+
- High-stakes decision making
|
63 |
+
- Generating factual content without verification
|
64 |
+
|
65 |
+
## How to Get Started
|
66 |
+
|
67 |
+
```python
|
68 |
+
from unsloth import FastLanguageModel
|
69 |
+
import torch
|
70 |
+
|
71 |
+
model, tokenizer = FastLanguageModel.from_pretrained(
|
72 |
+
model_name = "Srinivasmec26/MindSlate",
|
73 |
+
max_seq_length = 2048,
|
74 |
+
dtype = torch.float16,
|
75 |
+
load_in_4bit = True,
|
76 |
+
)
|
77 |
+
|
78 |
+
messages = [
|
79 |
+
{"role": "user", "content": "Create flashcards for neural networks:"},
|
80 |
+
]
|
81 |
+
|
82 |
+
inputs = tokenizer.apply_chat_template(
|
83 |
+
messages,
|
84 |
+
return_tensors = "pt",
|
85 |
+
).to("cuda")
|
86 |
+
|
87 |
+
outputs = model.generate(**inputs, max_new_tokens=256)
|
88 |
+
print(tokenizer.decode(outputs[0]))
|
89 |
+
```
|
90 |
+
|
91 |
+
## Training Details
|
92 |
+
|
93 |
+
### Training Data
|
94 |
+
|
95 |
+
- **Flashcards Dataset**: 400 items (cite your source)
|
96 |
+
- **Reminders Dataset**: 100 items (cite your source)
|
97 |
+
- **Summaries Dataset**: 100 items (cite your source)
|
98 |
+
- **Todos Dataset**: 100 items (cite your source)
|
99 |
+
|
100 |
+
*Replace with actual dataset citations and descriptions*
|
101 |
+
|
102 |
+
### Training Procedure
|
103 |
+
|
104 |
+
- **Preprocessing**: Standardized into "### Input: / ### Output:" format
|
105 |
+
- **Fine-tuned with**: Unsloth 2025.8.1
|
106 |
+
- **Hardware**: Tesla T4 GPU (16GB VRAM)
|
107 |
+
- **Training Time**: ~51 minutes for 3 epochs
|
108 |
+
- **LoRA Configuration**:
|
109 |
+
- Rank: 64
|
110 |
+
- Alpha: 128
|
111 |
+
- Target Modules: All key projection layers
|
112 |
+
|
113 |
+
## Evaluation
|
114 |
+
|
115 |
+
*Add evaluation metrics if available, for example:*
|
116 |
+
|
117 |
+
| Metric | Value |
|
118 |
+
|--------------|-------|
|
119 |
+
| Perplexity | X.XX |
|
120 |
+
| BLEU Score | X.XX |
|
121 |
+
| Training Loss| 0.128 |
|
122 |
+
|
123 |
+
## Technical Specifications
|
124 |
+
|
125 |
+
- **Model Size**: 3B parameters
|
126 |
+
- **Quantization**: 4-bit (bnb)
|
127 |
+
- **Context Length**: 2048 tokens
|
128 |
+
- **Precision**: bfloat16/fp16 mixed
|
129 |
+
|
130 |
+
## Citation
|
131 |
+
|
132 |
+
```bibtex
|
133 |
+
@misc{mindslate2025,
|
134 |
+
author = {Srinivas Nampalli},
|
135 |
+
title = {MindSlate: Fine-tuned Gemma-3B for Personal Knowledge Management},
|
136 |
+
year = {2025},
|
137 |
+
publisher = {Hugging Face},
|
138 |
+
howpublished = {\url{https://huggingface.co/Srinivasmec26/MindSlate}}
|
139 |
+
}
|
140 |
+
```
|
141 |
+
|
142 |
+
## Model Card Contact
|
143 |
+
|
144 |
+
For questions about MindSlate, contact:
|
145 |
+
- Srinivas Nampalli
|
146 |
+
- [LinkedIn](https://www.linkedin.com/in/srinivas-nampalli/)
|