MarshallDoyle commited on
Commit
8cd314b
·
verified ·
1 Parent(s): 9cfc350

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +55 -24
README.md CHANGED
@@ -1,49 +1,80 @@
1
  ---
2
  base_model: unsloth/meta-llama-3.1-8b-instruct-unsloth-bnb-4bit
3
  tags:
4
- - text-generation-inference
5
- - transformers
6
- - unsloth
7
- - llama
8
- - gguf
9
  license: apache-2.0
10
  language:
11
- - en
12
  ---
13
 
14
- # NASA Systems Engineering LLaMA-3.1 8B Instruct (Fine-Tuned)
15
 
16
- - **Developed by:** MarshallDoyle
17
- - **License:** Apache-2.0
18
- - **Fine-tuned from model:** unsloth/meta-llama-3.1-8b-instruct-unsloth-bnb-4bit
19
 
20
- This model specializes in answering questions and summarizing content from NASA’s *Systems Engineering Handbook* (2018) PDF. It was fine-tuned using Hugging Face’s TRL and LoRA techniques for high efficiency.
21
 
22
- ## Preview of Fine-Tuning Source
 
 
 
23
 
24
- ![Preview of NASA Systems Engineering Handbook first page](https://www.nasa.gov/sites/default/files/atoms/files/nasa_systems_engineering_handbook_0.pdf#page=1)
 
 
25
 
26
- *(Above: the front page of the NASA Systems Engineering Handbook used for fine-tuning.)*
 
 
 
 
 
27
 
28
  ---
29
 
30
- ## Overview
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
31
 
32
- **Key Features:**
 
33
 
34
- - **Domain Expertise:** Deep understanding of NASA systems engineering lifecycle phases, verification vs. validation testing, and SE best practices.
35
- - **Compact Footprint:** Keeps the lightweight 8B base model, adding only ~83.8M trainable parameters via LoRA.
36
- - **Extended Context:** Supports up to 128K tokens for long-form document Q&A.
 
37
 
38
- **Intended Use:** Ideal for educational tools, engineering Q&A assistants, mission planning support, and rapid handbook lookup.
39
 
40
  ---
41
 
42
- ## Model Card Updates
43
 
44
- - Removed Unsloth branding badges in favor of a direct preview of the NASA PDF source.
45
- - Added a representative image of the handbook’s first page to illustrate the training data.
 
46
 
47
  ---
48
 
49
- *
 
1
  ---
2
  base_model: unsloth/meta-llama-3.1-8b-instruct-unsloth-bnb-4bit
3
  tags:
4
+ - text-generation-inference
5
+ - transformers
6
+ - llama
7
+ - gguf
 
8
  license: apache-2.0
9
  language:
10
+ - en
11
  ---
12
 
13
+ # 🚀 NASA Systems Engineering LLaMA-3.1 8B Instruct (Fine-Tuned)
14
 
15
+ [![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
16
+ [![Parameters](https://img.shields.io/badge/Parameters-8.1B%20total%20(%2B83.8M%20LoRA)-lightgrey.svg)]()
17
+ [![Context Window](https://img.shields.io/badge/Context%20Window-128K%20tokens-lightgrey.svg)]()
18
 
19
+ ![NASA SE Handbook Preview](https://www.nasa.gov/sites/default/files/atoms/files/nasa_systems_engineering_handbook_0.pdf#page=1)
20
 
21
+ **Developed by:** MarshallDoyle
22
+ **Fine-tuned from:** `unsloth/meta-llama-3.1-8b-instruct-unsloth-bnb-4bit`
23
+ **Specialization:** NASA *Systems Engineering Handbook* (2018) Q&A
24
+ **Training recipe:** LoRA rank=16, epochs=3, steps=285, final loss≈0.40–0.60
25
 
26
+ ---
27
+
28
+ ## ✨ Key Features
29
 
30
+ | Feature | Details |
31
+ |-----------------------|--------------------------------------------------------|
32
+ | **Domain Expertise** | Deep knowledge of NASA SE lifecycle, terminology, and best practices. |
33
+ | **Compact Footprint** | +83.8M trainable via LoRA on 8B base. |
34
+ | **Long Context** | Up to 128K token support for large documents. |
35
+ | **Fast Inference** | Efficient TRL & Unsloth pipeline. |
36
 
37
  ---
38
 
39
+ ## 🎯 Intended Use Cases
40
+
41
+ - **Interactive Q&A:** Rapid lookup of definitions, processes, and phase purposes.
42
+ - **Summarization:** Concise overviews of complex sections in the handbook.
43
+ - **Educational Tools:** Training aids for systems engineering students and professionals.
44
+ - **Mission Planning:** Quick reference for verification, validation, and lifecycle checkpoints.
45
+
46
+ ---
47
+
48
+ ## 📊 Example Comparisons
49
+
50
+ <details>
51
+ <summary><strong>Phase C Purpose</strong></summary>
52
+
53
+ | Model | Response Snippet |
54
+ |-------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
55
+ | **Base LLaMA-3.1**| _“Phase C covers detailed design and preparations for manufacturing…”_ (generic) |
56
+ | **Fine-Tuned** | _“The main purpose is to complete the detailed design of system elements, fabricate hardware, and code software, ensuring all end products are defined for integration.”_ |
57
+
58
+ </details>
59
 
60
+ <details>
61
+ <summary><strong>Verification vs. Validation</strong></summary>
62
 
63
+ | Model | Response Snippet |
64
+ |-------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
65
+ | **Base LLaMA-3.1**| _“Verification checks requirements; validation checks user needs…”_ (too brief) |
66
+ | **Fine-Tuned** | _“Verification ensures the system meets specified requirements; validation confirms it satisfies stakeholder expectations in its intended environment.”_ |
67
 
68
+ </details>
69
 
70
  ---
71
 
72
+ ## ⚠️ Limitations
73
 
74
+ - **Scope Bounded:** Content based on the 2018 handbook; may omit newer updates.
75
+ - **Hallucinations:** Possible on out-of-handbook queries.
76
+ - **Bias:** NASA-specific terminology; verify critical engineering decisions against official documents.
77
 
78
  ---
79
 
80
+ 🔗 **Download & Details:** [Hugging Face model page](https://huggingface.co/MarshallDoyle/NASA-Systems-Engineering)