jackal79 commited on
Commit
86adbd9
·
verified ·
1 Parent(s): 88e1869

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +31 -33
README.md CHANGED
@@ -10,12 +10,9 @@ tags:
10
  license: other
11
  ---
12
 
13
-
14
  # tle-orbit-explainer
15
 
16
- A LoRA adapter for **Qwen-1.5-7B** that transforms raw Two-Line Elements (TLEs)
17
- into natural-language orbit explanations, decay risk scores, and anomaly flags
18
- for Space-Domain-Awareness (SDA) workflows.
19
 
20
  ---
21
 
@@ -23,23 +20,23 @@ for Space-Domain-Awareness (SDA) workflows.
23
 
24
  ### Model Description
25
 
26
- | | |
27
- |---|---|
28
- | **Developed by** | Jack Al-Kahwati / Stardrive |
29
- | **Funded by** | ⬜️ (Self-funded) |
30
- | **Shared by** | jackal79 (Hugging Face) |
31
- | **Model type** | LoRA adapter (`peft==0.10.0`) |
32
- | **Languages** | English |
33
- | **License** | apache-2.0 |
34
  | **Finetuned from** | [`Qwen/Qwen1.5-7B`](https://huggingface.co/Qwen/Qwen1.5-7B) |
35
 
36
  ### Model Sources
37
 
38
- | | |
39
- |---|---|
40
- | **Repository** | <https://huggingface.co/jackal79/tle-orbit-explainer> |
41
- | **Paper / Blog** | ⬜️ (optional link) |
42
- | **Demo** | ⬜️ Gradio / Space (optional) |
43
 
44
  ---
45
 
@@ -47,36 +44,36 @@ for Space-Domain-Awareness (SDA) workflows.
47
 
48
  ### Direct Use
49
 
50
- * Rapid orbital-state summarisation for flight-dynamics teams
51
- * Analyst chat-assistants that translate TLEs into plain English
52
- * Offline dataset annotation (adding orbit-class labels)
53
 
54
  ### Downstream Use
55
 
56
- * Fuse with SGP4 for full position forecasting
57
- * Embed in on-board autonomy stacks
58
- * Pre-prompted agent in secure pipelines
59
 
60
  ### Out-of-Scope Use
61
 
62
- * Precise orbit propagation without a physics engine
63
- * Weapon-targeting or lethal-autonomy decision loops
64
- * Any jurisdiction that prohibits ML export (check ITAR/EAR)
65
 
66
  ---
67
 
68
  ## Bias, Risks, & Limitations
69
 
70
- | Category | Note |
71
- |----------|------|
72
- | **Data bias** | Trained only on *decayed* objects (`DECAY = 1`) may under-predict longevity for active constellations. |
73
- | **Temporal limits** | Snapshot reasoning; no 1 Hz time-series learned yet. |
74
- | **Language** | English explanations only. |
75
- | **Security** | Model could hallucinate wrong decay dates always cross-check. |
76
 
77
  ### Recommendations
78
 
79
- Integrate physics-based checks before acting on decay predictions; keep human-in-the-loop for any safety-critical task.
80
 
81
  ---
82
 
@@ -102,3 +99,4 @@ prompt = """### Prompt:
102
  ### Reasoning:
103
  """
104
  print(pipe(prompt, max_new_tokens=120)[0]["generated_text"])
 
 
10
  license: other
11
  ---
12
 
 
13
  # tle-orbit-explainer
14
 
15
+ A LoRA adapter for **Qwen-1.5-7B** that translates raw Two-Line Elements (TLEs) into natural-language orbit explanations, decay risk scores, and anomaly flags for general space awareness workflows.
 
 
16
 
17
  ---
18
 
 
20
 
21
  ### Model Description
22
 
23
+ | | |
24
+ | ------------------ | ----------------------------------------------------------- |
25
+ | **Developed by** | Jack Al-Kahwati / Stardrive |
26
+ | **Funded by** | ⬜️ (Self-funded) |
27
+ | **Shared by** | jackal79 (Hugging Face) |
28
+ | **Model type** | LoRA adapter (`peft==0.10.0`) |
29
+ | **Languages** | English |
30
+ | **License** | apache-2.0 |
31
  | **Finetuned from** | [`Qwen/Qwen1.5-7B`](https://huggingface.co/Qwen/Qwen1.5-7B) |
32
 
33
  ### Model Sources
34
 
35
+ | | |
36
+ | ---------------- | ---------------------------------------------------------------------------------------------------------- |
37
+ | **Repository** | [https://huggingface.co/jackal79/tle-orbit-explainer](https://huggingface.co/jackal79/tle-orbit-explainer) |
38
+ | **Paper / Blog** | ⬜️ (optional link) |
39
+ | **Demo** | ⬜️ Gradio / Space (optional) |
40
 
41
  ---
42
 
 
44
 
45
  ### Direct Use
46
 
47
+ * Quick summarization of satellite orbital states for analysts
48
+ * Plain-language TLE explanations for educational purposes
49
+ * Offline dataset labeling (orbital classifications)
50
 
51
  ### Downstream Use
52
 
53
+ * Combine with SGP4 for enhanced position forecasting
54
+ * Integration into satellite autonomy stacks (cubesats, small-scale hardware)
55
+ * Pre-prompted agent support in secure orbital management workflows
56
 
57
  ### Out-of-Scope Use
58
 
59
+ * High-precision orbit propagation without additional physics modeling
60
+ * Applications related to targeting, weapons systems, or lethal autonomous decisions
61
+ * Jurisdictions prohibiting ML or data export (verify with ITAR/EAR guidelines)
62
 
63
  ---
64
 
65
  ## Bias, Risks, & Limitations
66
 
67
+ | Category | Note |
68
+ | ------------------- | ------------------------------------------------------------------------------------------------------------- |
69
+ | **Data bias** | Trained primarily on decayed objects (`DECAY = 1`), possibly underestimating longevity for active satellites. |
70
+ | **Temporal limits** | Operates on snapshot data; does not handle continuous high-frequency time-series. |
71
+ | **Language** | Supports explanations in English only. |
72
+ | **Accuracy** | Potential inaccuracies in decay date predictions; verify independently. |
73
 
74
  ### Recommendations
75
 
76
+ Incorporate independent physics-based validation before operational use and maintain a human-in-the-loop for any critical or high-risk decisions.
77
 
78
  ---
79
 
 
99
  ### Reasoning:
100
  """
101
  print(pipe(prompt, max_new_tokens=120)[0]["generated_text"])
102
+ ```