tle-orbit-explainer / README.md
jackal79's picture
Update README.md
f5649bc verified
|
raw
history blame
2.87 kB
metadata
base_model: Qwen/Qwen1.5-7B
library_name: peft
tags:
  - LoRA
  - TLE
  - space-domain-awareness
  - trajectory-prediction
  - orbital-mechanics
license: other

tle-orbit-explainer

A LoRA adapter for Qwen-1.5-7B that transforms raw Two-Line Elements (TLEs) into natural-language orbit explanations, decay risk scores, and anomaly flags for Space-Domain-Awareness (SDA) workflows.


Model Details

Model Description

Developed by Jack Al-Kahwati / Stardrive
Funded by ⬜️ (Self-funded)
Shared by jackal79 (Hugging Face)
Model type LoRA adapter (peft==0.10.0)
Languages English
License apache-2.0
Finetuned from Qwen/Qwen1.5-7B

Model Sources

Repository https://huggingface.co/jackal79/tle-orbit-explainer
Paper / Blog ⬜️ (optional link)
Demo ⬜️ Gradio / Space (optional)

Uses

Direct Use

  • Rapid orbital-state summarisation for flight-dynamics teams
  • Analyst chat-assistants that translate TLEs into plain English
  • Offline dataset annotation (adding orbit-class labels)

Downstream Use

  • Fuse with SGP4 for full position forecasting
  • Embed in on-board autonomy stacks (cubesats, ESP-32 class)
  • Pre-prompted agent in secure SDA pipelines (Space Force, SDA, JSpOC)

Out-of-Scope Use

  • Precise orbit propagation without a physics engine
  • Weapon-targeting or lethal-autonomy decision loops
  • Any jurisdiction that prohibits ML export (check ITAR/EAR)

Bias, Risks, & Limitations

Category Note
Data bias Trained only on decayed objects (DECAY = 1) → may under-predict longevity for active constellations.
Temporal limits Snapshot reasoning; no 1 Hz time-series learned yet.
Language English explanations only.
Security Model could hallucinate wrong decay dates → always cross-check.

Recommendations

Integrate physics-based checks before acting on decay predictions; keep human-in-the-loop for any safety-critical task.


How to Get Started

from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from peft import PeftModel

base = "Qwen/Qwen1.5-7B"
lora = "jackal79/tle-orbit-explainer"

tok   = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, device_map="auto")
model = PeftModel.from_pretrained(model, lora)  # merges LoRA

pipe = pipeline("text-generation", model=model, tokenizer=tok, device=0)

prompt = """### Prompt:
1 25544U 98067A   24079.07757601 .00016717 00000+0 10270-3 0  9994
2 25544  51.6400 337.6640 0007776  35.5310 330.5120 15.50377579499263

### Reasoning:
"""
print(pipe(prompt, max_new_tokens=120)[0]["generated_text"])