Update README.md
#9
by
Taylor658
- opened
README.md
CHANGED
|
@@ -5,115 +5,100 @@ datasets:
|
|
| 5 |
language:
|
| 6 |
- en
|
| 7 |
base_model:
|
| 8 |
-
- mistralai/
|
| 9 |
tags:
|
| 10 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
- hohmann-transfer-orbits
|
| 12 |
-
- text-generation-inference
|
| 13 |
-
- transformers
|
| 14 |
---
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
**Model Type**: Transformer-based language model
|
| 24 |
-
**Languages**: English
|
| 25 |
-
**License**: MIT
|
| 26 |
-
|
| 27 |
-
This model is based on the **Mistral-8x7B** foundation model and is being fine tuned on the **titan-hohmann-transfer-orbit** dataset. It is designed to assist with advanced orbital calculations, specifically focusing on space probe Hohmann transfer orbits, and includes support for native function calling and multimodal inputs.
|
| 28 |
-
|
| 29 |
-
---
|
| 30 |
-
|
| 31 |
-
### Model Details
|
| 32 |
-
|
| 33 |
-
- **Developers**: A Taylor
|
| 34 |
-
- **Model Architecture**: Transformer based with enhancements for code generation and multimodal processing
|
| 35 |
-
- **Parameters**: 7 Billion
|
| 36 |
-
- **Native Function Calling**: Supported
|
| 37 |
-
- **Multimodal Capabilities**: Text and Image processing
|
| 38 |
-
|
| 39 |
-
---
|
| 40 |
-
|
| 41 |
-
### Intended Use
|
| 42 |
-
|
| 43 |
-
- **Primary Applications**:
|
| 44 |
-
- Assist aerospace engineers with calculations related to Hohmann transfer orbits
|
| 45 |
-
- Facilitate research by offering computational assistance in orbital dynamics
|
| 46 |
-
- **Usage Scenarios**:
|
| 47 |
-
- Calculating delta-v requirements
|
| 48 |
-
- Estimating transfer times between celestial bodies
|
| 49 |
-
- Visualizing orbital trajectories
|
| 50 |
-
|
| 51 |
-
---
|
| 52 |
-
|
| 53 |
-
### Training Data
|
| 54 |
-
|
| 55 |
-
- **Dataset Name**: `titan-hohmann-transfer-orbit`
|
| 56 |
-
- **Description**: A specialized dataset containing textual explanations, code snippets, and diagrams related to probe orbital mechanics and Hohmann transfer orbits.
|
| 57 |
-
- **Data Modalities**:
|
| 58 |
-
- **Text**: Technical documentation and literature on orbital mechanics
|
| 59 |
-
- **Code**: Sample code for orbital calculations and simulations
|
| 60 |
-
- **Images**: Diagrams illustrating probe orbital paths and spacecraft maneuvers
|
| 61 |
-
|
| 62 |
---
|
| 63 |
|
| 64 |
-
|
| 65 |
|
| 66 |
-
|
| 67 |
|
| 68 |
-
|
| 69 |
-
2. **Native Function Calling**: Integrating the ability to execute predefined computational functions within responses for accurate and dynamic calculations.
|
| 70 |
-
3. **Domain-Specific Fine-Tuning**: Refining the model's parameters with the titan-hohmann-transfer-orbit dataset to enhance accuracy and expertise in calculating Hohmann transfer trajectories.
|
| 71 |
-
4. **Validation and Testing**: Verifying the model outputs against industry standards and benchmarks to ensure precision and reliability in aerospace applications.
|
| 72 |
|
| 73 |
-
|
| 74 |
|
| 75 |
-
|
| 76 |
|
| 77 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
|
| 79 |
-
|
| 80 |
|
| 81 |
-
|
| 82 |
-
|
|
|
|
| 83 |
|
| 84 |
-
|
| 85 |
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
|
|
|
| 89 |
|
| 90 |
-
|
|
|
|
| 91 |
|
| 92 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
|
| 94 |
-
|
| 95 |
-
|
|
|
|
|
|
|
| 96 |
|
| 97 |
-
|
|
|
|
|
|
|
| 98 |
|
| 99 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
|
| 101 |
-
|
| 102 |
-
- **Dataset Contributors**: A Taylor
|
| 103 |
-
- **Open-Source Community**: Appreciation for the tools and libraries that made this project possible.
|
| 104 |
|
| 105 |
-
|
| 106 |
-
|
| 107 |
-
### License
|
| 108 |
|
| 109 |
-
|
| 110 |
-
- **Dataset License**: MIT
|
| 111 |
|
| 112 |
-
|
|
|
|
| 113 |
|
| 114 |
-
|
| 115 |
|
| 116 |
-
|
|
|
|
| 117 |
|
| 118 |
### Contact Information
|
| 119 |
|
|
|
|
| 5 |
language:
|
| 6 |
- en
|
| 7 |
base_model:
|
| 8 |
+
- mistralai/Pixtral-12B-Base-2409
|
| 9 |
tags:
|
| 10 |
+
- mistral
|
| 11 |
+
- pixtral
|
| 12 |
+
- vlm
|
| 13 |
+
- multimodal
|
| 14 |
+
- image-text-to-text
|
| 15 |
+
- orbital-mechanics
|
| 16 |
- hohmann-transfer-orbits
|
|
|
|
|
|
|
| 17 |
---
|
| 18 |
+
language:
|
| 19 |
+
- en
|
| 20 |
+
license: mit
|
| 21 |
+
library_name: transformers
|
| 22 |
+
pipeline_tag: image-text-to-text
|
| 23 |
+
base_model: mistralai/Pixtral-12B-Base-2409
|
| 24 |
+
datasets:
|
| 25 |
+
- Taylor658/titan-hohmann-transfer-orbit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
---
|
| 27 |
|
| 28 |
+
# Pixtral 12B Fine-Tuned on Titan-Hohmann-Transfer-Orbit
|
| 29 |
|
| 30 |
+
> Updated to the latest suitable Mistral multimodal base: `mistralai/Pixtral-12B-Base-2409`.
|
| 31 |
|
| 32 |
+
## Overview
|
|
|
|
|
|
|
|
|
|
| 33 |
|
| 34 |
+
Fine-tuned variant of Pixtral 12B for orbital mechanics with emphasis on Hohmann transfer orbits. Supports multimodal (image + text) inputs and text outputs.
|
| 35 |
|
| 36 |
+
## Model Details
|
| 37 |
|
| 38 |
+
- Base: `mistralai/Pixtral-12B-Base-2409`
|
| 39 |
+
- Type: Multimodal (Vision + Text)
|
| 40 |
+
- Params: ~12B (decoder) + vision encoder
|
| 41 |
+
- Languages: English
|
| 42 |
+
- License: MIT
|
| 43 |
|
| 44 |
+
## Intended Use
|
| 45 |
|
| 46 |
+
- Hohmann transfer ∆v estimation
|
| 47 |
+
- Transfer-time approximations
|
| 48 |
+
- Orbit analysis aids and reasoning
|
| 49 |
|
| 50 |
+
## Quickstart
|
| 51 |
|
| 52 |
+
### vLLM (multimodal)
|
| 53 |
+
```python
|
| 54 |
+
from vllm import LLM
|
| 55 |
+
from vllm.sampling_params import SamplingParams
|
| 56 |
|
| 57 |
+
llm = LLM(model="mistralai/Pixtral-12B-Base-2409", tokenizer_mode="mistral")
|
| 58 |
+
sampling = SamplingParams(max_tokens=512, temperature=0.2)
|
| 59 |
|
| 60 |
+
messages = [
|
| 61 |
+
{
|
| 62 |
+
"role": "user",
|
| 63 |
+
"content": [
|
| 64 |
+
{"type": "text", "text": "Given this diagram, estimate the delta-v for a Hohmann transfer to Titan."},
|
| 65 |
+
{"type": "image_url", "image_url": {"url": "https://example.com/orbit_diagram.png"}}
|
| 66 |
+
]
|
| 67 |
+
}
|
| 68 |
+
]
|
| 69 |
+
resp = llm.chat(messages, sampling_params=sampling)
|
| 70 |
+
print(resp[0].outputs[0].text)
|
| 71 |
+
```
|
| 72 |
|
| 73 |
+
### Transformers (text-only demo)
|
| 74 |
+
```python
|
| 75 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 76 |
+
import torch
|
| 77 |
|
| 78 |
+
model_id = "mistralai/Pixtral-12B-Base-2409"
|
| 79 |
+
tok = AutoTokenizer.from_pretrained(model_id)
|
| 80 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto", device_map="auto")
|
| 81 |
|
| 82 |
+
prompt = "Compute approximate delta-v for a Hohmann transfer to Titan. State assumptions."
|
| 83 |
+
inputs = tok(prompt, return_tensors="pt").to(model.device)
|
| 84 |
+
out = model.generate(**inputs, max_new_tokens=512, temperature=0.2)
|
| 85 |
+
print(tok.decode(out[0], skip_special_tokens=True))
|
| 86 |
+
```
|
| 87 |
|
| 88 |
+
## Training Data
|
|
|
|
|
|
|
| 89 |
|
| 90 |
+
- Dataset: `Taylor658/titan-hohmann-transfer-orbit`
|
| 91 |
+
- Modalities: text (explanations), code (snippets), images (orbital diagrams)
|
|
|
|
| 92 |
|
| 93 |
+
## Limitations
|
|
|
|
| 94 |
|
| 95 |
+
- Optimized for Hohmann transfers and related reasoning
|
| 96 |
+
- Requires sufficient GPU VRAM for best throughput
|
| 97 |
|
| 98 |
+
## Acknowledgements
|
| 99 |
|
| 100 |
+
- Base model by Mistral AI (Pixtral 12B)
|
| 101 |
+
- Dataset by A Taylor
|
| 102 |
|
| 103 |
### Contact Information
|
| 104 |
|