Hugging Face's logo

Inserloft
/
NaNo

Text Generation
Transformers
GGUF
English
ai
llm
edge-ai
mobile-ai
code
programming
lightweight
inserloft
nano
Model card Files
xet
 Community
Inserloft commited on
Commit
8d59df9
·
verified ·
1 Parent(s): 714219d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +167 -21
README.md CHANGED
@@ -1,36 +1,182 @@
1
  ---
2
  language:
3
- - es
4
  - en
5
  license: mit
6
  tags:
7
- - gpt2
 
 
 
8
  - code
9
- - bilingual
 
10
  - inserloft
11
- model_name: Cleo Nano v3.1 Bilingual
 
 
 
12
  ---
13
 
14
- # Cleo Nano v3.1 (Bilingual Optimization)
15
 
16
- Cleo Nano is a decoder-only Transformer model developed by **Inserloft** under the vision of **Jesus Heriberto Corona**. This version (v3.1) features surgical fine-tuning for bilingual stability (English/Spanish) and hallucination control.
17
 
18
- ## Model Details
19
- - **Architecture:** Decoder-Only GPT (Custom)
20
- - **Layers:** 8
21
- - **Embedding Dim:** 384
22
- - **Attention Heads:** 12
23
- - **Context Window:** 256 tokens
24
- - **Parameters:** ~15M
25
- - **Training Data:** Mix of Wikipedia, Python Code (CodeFeedback), and Identity Anchoring.
26
 
27
- ## Usage
28
- To use this model, you need the custom `CleoNanoV3` architecture defined in PyTorch. The weights can be loaded using `torch.load()` or via the Hugging Face `from_pretrained` if using the provided mapping logic.
29
 
30
- ### Capabilities
31
- 1. **Bilingual Chat:** Responds to general queries in both Spanish and English.
32
- 2. **Code Generation:** Specialized in Python snippets (Sum, Loops, Classes).
33
- 3. **Identity Preservation:** Strong grounding on its origin and creator.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
34
 
35
  ---
36
- Developed by [Inserloft](https://inserloft.dev/)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  language:
 
3
  - en
4
  license: mit
5
  tags:
6
+ - ai
7
+ - llm
8
+ - edge-ai
9
+ - mobile-ai
10
  - code
11
+ - programming
12
+ - lightweight
13
  - inserloft
14
+ - nano
15
+ pipeline_tag: text-generation
16
+ library_name: transformers
17
+ model_name: NaNo 3.1
18
  ---
19
 
20
+ # NaNo 3.1
21
 
22
+ NaNo 3.1 is a lightweight AI language model developed by Inserloft, designed primarily for programming, edge AI, mobile inference, and efficient local deployment.
23
 
24
+ Unlike large-scale general-purpose models, NaNo focuses on delivering strong technical and coding-oriented capabilities while maintaining low resource consumption and fast inference speeds.
 
 
 
 
 
 
 
25
 
26
+ NaNo is part of the broader Inserloft AI ecosystem alongside larger and more advanced models such as Kyro.
 
27
 
28
+ ---
29
+
30
+ # Overview
31
+
32
+ NaNo was built around a simple philosophy:
33
+
34
+ > Efficient AI models should be capable, fast, lightweight, and deployable almost anywhere.
35
+
36
+ NaNo 3.1 introduces major improvements in:
37
+ - Context handling
38
+ - Technical reasoning
39
+ - Programming capabilities
40
+ - Conversational stability
41
+ - Inference optimization
42
+ - Deployment efficiency
43
+
44
+ This version also represents the largest scaling upgrade in the model family so far.
45
+
46
+ ---
47
+
48
+ # What's New in NaNo 3.1
49
+
50
+ ## Major Parameter Scaling
51
+
52
+ NaNo 3.1 scales from:
53
+ - **22M → 52M parameters**
54
+
55
+ This significant increase improves:
56
+ - Code understanding
57
+ - Response coherence
58
+ - Technical reasoning
59
+ - Long-context retention
60
+ - Structured generation quality
61
+
62
+ while preserving NaNo's lightweight deployment philosophy.
63
+
64
+ ---
65
+
66
+ # Core Focus Areas
67
+
68
+ ## Programming
69
+
70
+ NaNo is heavily optimized for:
71
+ - Code generation
72
+ - Function completion
73
+ - Technical assistance
74
+ - Refactoring
75
+ - Automation workflows
76
+ - Structured programming tasks
77
+
78
+ ---
79
+
80
+ ## Edge AI
81
+
82
+ NaNo is designed for modern edge computing environments:
83
+ - Lightweight servers
84
+ - Embedded systems
85
+ - Local AI applications
86
+ - Edge devices
87
+ - Efficient hardware deployment
88
+
89
+ ---
90
+
91
+ ## Mobile AI
92
+
93
+ NaNo prioritizes:
94
+ - Fast inference
95
+ - Lower memory usage
96
+ - Mobile compatibility
97
+ - On-device execution
98
+ - Offline AI experiences
99
+
100
+ ---
101
+
102
+ # Model Details
103
+
104
+ | Category | Value |
105
+ |---|---|
106
+ | Architecture | Decoder-Only Transformer |
107
+ | Model Family | NaNo |
108
+ | Version | 3.1 |
109
+ | Parameters | ~52M |
110
+ | Primary Focus | Programming & Edge AI |
111
+ | Deployment Target | Mobile, Local, Edge |
112
+ | License | MIT |
113
 
114
  ---
115
+
116
+ # Technical Improvements
117
+
118
+ NaNo 3.1 includes improvements across:
119
+ - Attention stability
120
+ - Context retention
121
+ - Technical instruction following
122
+ - Code consistency
123
+ - Generation quality
124
+ - Inference optimization
125
+
126
+ The model is specifically optimized for technical and programming-oriented workflows rather than broad educational or general-purpose assistant behavior.
127
+
128
+ ---
129
+
130
+ # Inserloft AI Ecosystem
131
+
132
+ NaNo is part of the AI ecosystem developed by Inserloft.
133
+
134
+ Current model ecosystem:
135
+ - **NaNo** → Lightweight programming and edge AI
136
+ - **Kyro** → Advanced large-scale reasoning and intelligence
137
+
138
+ This specialization allows each model family to focus on specific real-world use cases.
139
+
140
+ ---
141
+
142
+ # Intended Use Cases
143
+
144
+ NaNo is intended for:
145
+ - Coding assistants
146
+ - Local AI tools
147
+ - Mobile AI systems
148
+ - Edge AI applications
149
+ - Lightweight inference environments
150
+ - Embedded AI workflows
151
+
152
+ ---
153
+
154
+ # Future Development
155
+
156
+ Future NaNo versions are expected to include:
157
+ - Longer context windows
158
+ - Better multilingual support
159
+ - Improved reasoning
160
+ - Faster inference
161
+ - Better code generation
162
+ - Mobile-specific optimizations
163
+ - More efficient architectures
164
+
165
+ ---
166
+
167
+ # Disclaimer
168
+
169
+ NaNo is an actively evolving experimental AI model.
170
+
171
+ Outputs may still contain inaccuracies, hallucinations, or unstable generations depending on prompts, deployment environments, and inference configurations.
172
+
173
+ ---
174
+
175
+ # Links
176
+
177
+ - Website: https://inserloft.dev
178
+ - Hugging Face Organization: https://huggingface.co/Inserloft
179
+
180
+ ---
181
+
182
+ Developed by Inserloft.