Update README.md

README.md

@@ -3,20 +3,164 @@ base_model: unsloth/gpt-oss-20b-unsloth-bnb-4bit
 tags:
 - text-generation-inference
 - transformers
-- unsloth
 - gpt_oss
 - trl
+- nasa
+- standards
 license: apache-2.0
 language:
 - en
 ---
 
-# Uploaded  model
+# NASA OSS Model Card
 
-- **Developed by:** MarshallDoyle
-- **License:** apache-2.0
-- **Finetuned from model :** unsloth/gpt-oss-20b-unsloth-bnb-4bit
+## Highlights
 
-This gpt_oss model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
+- Fine-tune of **OpenAI GPT-OSS** (20B) using [Unsloth](https://github.com/unslothai/unsloth) for optimized training.  
+- Trained on **synthetic Q&A data** derived from all available NASA standards and handbooks (excluding center-level standards).  
+- Data generated via **chunking into 4096 tokens with 256 overlap**, question + answer pairs produced per chunk.  
+- Provides **compliance-oriented, clause-referenced outputs** for NASA engineering standards.  
+- Extends prior work ([NASA Systems Engineering Llama-3.1 8B](https://huggingface.co/MarshallDoyle/NASA-Systems-Engineering)) to **dozens of NASA standards**, not just one handbook.  
+- **Best used with retrieval-augmented generation (RAG)**: include the relevant standard text in the prompt for highest accuracy.  
 
-[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
+**Recommended Inference Settings:**  
+- `temperature = 1`  
+- `top_k = 0`  
+- `top_p = 1`  
+
+**Official OpenAI GPT-OSS page:** [https://openai.com/oss](https://openai.com/oss)
+
+---
+
+## Model Summary
+
+The **NASA OSS model** is designed to assist with **interpretation, reasoning, and compliance guidance across NASA standards and handbooks**.  
+It is particularly effective for:  
+
+- Engineering decision support  
+- Standards compliance verification  
+- Requirements and process traceability  
+- Education and training of new engineers  
+
+---
+
+## Training Data
+
+- **Sources:** All publicly available NASA standards and handbooks (excluding center-level standards).  
+- **Processing pipeline:**  
+  - Documents chunked into 4096 tokens with 256-token overlap.  
+  - Each chunk used to generate multiple **synthetic questions** relevant only to that section.  
+  - **Answers generated** strictly from the chunk (but phrased naturally without explicit chunk references).  
+
+---
+
+## Recommended Usage
+
+- **Use with RAG:** Retrieval of the intended NASA standard or handbook section and injection into the prompt significantly improves performance.  
+- **Verification:** Always cross-check outputs against the official NASA source documents before applying in mission-critical work.  
+
+---
+
+## Comparison Examples
+
+### Example 1 – Required Components for Material Process Qualification (NASA-STD-6030)
+
+**Llama-3.1 8B Fine-tune (Systems Eng only):**  
+"Nasa standards generally require process control and material specifications before flight qualification."  
+
+**GPT-OSS Base:**  
+"Material specifications and process definitions must be established for qualification."  
+
+**NASA OSS:**  
+NASA-STD-6030 – Materials Process Qualification  
+The standard requires that a candidate material process be fully characterized before it can be qualified for use in additive manufacturing hardware. Three distinct components must be defined and documented:  
+1. Feedstock Material Specification – properties of the raw AM material.  
+2. Part Material Specification (or equivalent) – final part-material characteristics after processing.  
+3. Material Processing Specification (MPS) – process parameters and QA procedures.  
+These items are listed in NASA-STD-6030 §2 f (g–i) and clarified in §8 as required for a Qualified Part Process.  
+
+---
+
+### Example 2 – PCRDs (Primary Control-Relevant Data) for Metallic AM Processes
+
+**Llama-3.1 8B Fine-tune:**  
+"Typical values include tensile strength and elongation."  
+
+**GPT-OSS Base:**  
+"PCRDs are mechanical properties but fracture toughness is sometimes excluded."  
+
+**NASA OSS:**  
+Question: Which of the following is NOT one of the four PCRDs defined for each metallic AM MPS: ultimate tensile strength, yield strength, elongation, or fracture toughness?  
+Answer: Fracture toughness is NOT one of the PCRDs.  
+NASA-STD-6030 explicitly defines: ultimate tensile strength, yield strength, elongation, and reduction of area.  
+Reference: NASA-STD-6030 §4.3.2  
+
+---
+
+### Example 3 – Powder Certification and Reuse Rules for Ti-6Al-4V (LPBF)
+
+**Llama-3.1 8B Fine-tune:**  
+"Powder must be certified before use in additive manufacturing."  
+
+**GPT-OSS Base:**  
+"Certification and reuse requirements depend on program standards."  
+
+**NASA OSS:**  
+NASA-STD-6030 requires:  
+- Feedstock certification for composition, cleanliness, and particle morphology (§5.1).  
+- Documentation of lot traceability and maximum reuse cycles (§5.2).  
+- Validation via witness coupon testing before production (§7.3).  
+These steps ensure powder consistency and qualification evidence for flight hardware.  
+
+---
+
+### Example 4 – Wire-Feed Directed Energy Deposition (DED) Qualification
+
+**NASA OSS:**  
+For wire-feed DED processes, NASA-STD-6030 mandates documentation of:  
+- Material Specification – composition, purity, heat-treatment requirements (§7.2).  
+- Processing Parameters – machine-specific build settings and post-processing (§7.3).  
+- Testing & Inspection Methods – destructive and nondestructive evaluations with acceptance criteria (§7.3).  
+These form the Candidate Material Process (CMP) and serve as the foundation for establishing a Qualified Part Process (QPP).  
+
+---
+
+### Example 5 – Dimensional Inspection for AM Structural Truss
+
+**NASA OSS:**  
+NASA-STD-6030 requires GD&T compliance verification and interface checks through the Additive Manufacturing Control Plan (AMCP).  
+- MPS, QMP, and AMCP integration define dimensional verification (§4.1–4.3).  
+- Witness coupon testing and statistical sampling ensure dimensional repeatability (§7.2–7.3).  
+Reference: NASA-STD-6030, §4.2; §7.2–7.3  
+
+---
+
+## Limitations
+
+- Model outputs reflect **public NASA standards only**.  
+- May not cover internal center-level or proprietary standards.  
+- **Best used with retrieval context** – performance drops without standard text injection.  
+
+---
+
+## Ethical Considerations
+
+- Should be treated as an **assistive tool**, not as a replacement for human engineering judgment.  
+- Outputs must be verified against authoritative NASA documentation.  
+- Not suitable for export-controlled, ITAR-restricted, or classified projects.  
+
+---
+
+## Citation
+
+If you use this model, please cite as:
+
+@misc{marshall2025nasaoss,
+  author = {Marshall Doyle},
+  title = {NASA OSS: Domain-Specific Fine-Tune of GPT OSS on NASA Standards},
+  year = {2025},
+  publisher = {Hugging Face},
+  howpublished = {\url{https://huggingface.co/MarshallDoyle/NASA-OSS}}
+}
+
+---