|
--- |
|
license: other |
|
license_name: nakshatra-license |
|
license_link: LICENSE |
|
pipeline_tag: text-generation |
|
language: |
|
- en |
|
tags: |
|
- Nakshatra |
|
base_model: |
|
- OEvortex/HelpingAI2-6B |
|
--- |
|
|
|
# Nakshatra: Human-like Conversational AI Prototype |
|
|
|
 |
|
|
|
## Overview |
|
Nakshatra is a groundbreaking prototype AI model, boasting **10x** better human-like responses compared to the previous HelpingAI models. Designed by **Abhay Koul (OEvortex)**, Nakshatra leverages advanced conversational techniques to deliver highly coherent, empathetic, and contextually aware interactions, making it a major leap forward in AI-human interaction. |
|
|
|
- Delivers near-human conversational quality and responsiveness.- Delivers near-human conversational quality and responsiveness. |
|
- Exhibits deep contextual understanding and emotional intelligence in interactions. |
|
- Aimed at providing more natural, emotionally intuitive dialogue experiences.- Aimed at providing more natural, emotionally intuitive dialogue experiences. |
|
|
|
## Methodology |
|
Nakshatra employs a combination of the following techniques to achieve its remarkable conversational capabilities: |
|
- **Supervised Learning**: Trained with vast dialogue datasets, including those with emotional annotations, to ensure it can handle a wide range of conversational contexts. |
|
- **Human-like Conversation Training**: Fine-tuned to imitate natural human conversational patterns. |
|
- **Prototype Optimization**: This version is still in the prototype phase but showcases significant advancements in language coherence, tone, and emotional sensitivity. |
|
|
|
## Usage Code |
|
```python |
|
import torch |
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
|
# Load the Nakshatra model |
|
model = AutoModelForCausalLM.from_pretrained("OEvortex/Nakshatra", trust_remote_code=True) |
|
# Load the tokenizer |
|
tokenizer = AutoTokenizer.from_pretrained("OEvortex/Nakshatra", trust_remote_code=True) |
|
|
|
# Define the chat input |
|
chat = [ |
|
{ "role": "system", "content": "You are Nakshatra, a human-like conversational AI. Answer in the most human way possible." }, |
|
{ "role": "user", "content": "Introduce yourself!" } |
|
] |
|
|
|
inputs = tokenizer.apply_chat_template( |
|
chat, |
|
add_generation_prompt=True, |
|
return_tensors="pt" |
|
).to(model.device) |
|
|
|
# Generate text |
|
outputs = model.generate( |
|
inputs, |
|
max_new_tokens=256, |
|
do_sample=True, |
|
temperature=0.6, |
|
top_p=0.9, |
|
eos_token_id=tokenizer.eos_token_id |
|
) |
|
|
|
response = outputs[0][inputs.shape[-1]:] |
|
print(tokenizer.decode(response, skip_special_tokens=True)) |
|
``` |
|
|
|
## Using the Model with GGUF |
|
```python |
|
# %pip install -U 'webscout[local]' -q |
|
|
|
from webscout.Local.utils import download_model |
|
from webscout.Local.model import Model |
|
from webscout.Local.thread import Thread |
|
from webscout.Local import formats |
|
from webscout.Local.samplers import SamplerSettings |
|
|
|
# Download the model |
|
repo_id = "OEvortex/Nakshatra" |
|
filename = "nakshatra-q4_k_m.gguf" |
|
model_path = download_model(repo_id, filename, token=None) |
|
|
|
# Load the model |
|
model = Model(model_path, n_gpu_layers=40) |
|
|
|
# Define the system prompt |
|
system_prompt = "You are Nakshatra, a human-like conversational AI. Answer in the most human way possible." |
|
|
|
# Create a chat format with your system prompt |
|
nakshatra_format = formats.llama3.copy() |
|
nakshatra_format['system_content'] = system_prompt |
|
|
|
# Define your sampler settings (optional) |
|
sampler = SamplerSettings(temp=0.7, top_p=0.9) |
|
|
|
# Create a Thread with the custom format and sampler |
|
thread = Thread(model, nakshatra_format, sampler=sampler) |
|
|
|
# Start interacting with the model |
|
thread.interact(header="🌟 Nakshatra - Human-like AI Prototype 🚀", color=True) |
|
``` |
