metadata

tags:
  - text-generation-inference
  - text-generation
  - Sentiment Analysis
  - qlora
  - peft
license: apache-2.0
library_name: transformers
widget:
  - messages:
      - role: user
        content: What is your name?
language:
  - en
  - ro
pipeline_tag: text-generation
model-index:
  - name: CognitivessAI/cognitivess
    results:
      - task:
          type: text-generation
          name: Text Generation
        metrics:
          - name: Evaluation Status
            type: accuracy
            value: Pending
            description: >-
              Comprehensive evaluations are planned and will be conducted in the
              future.
model_type: CognitivessForCausalLM
quantization_config:
  load_in_8bit: true
  llm_int8_threshold: 6
fine_tuning:
  method: qlora
  peft_type: LORA
inference:
  parameters:
    max_new_tokens: 8192
    temperature: 0.7
    top_p: 0.95
    do_sample: true

Accessible and portable generative AI solutions for developers and businesses.

Website | Demo | GitHub

Cognitivess

Cognitivess is an advanced language model developed by Cognitivess AI, based in Bucharest, Romania. This model is trained from scratch on a diverse and curated dataset, encompassing a wide range of knowledge domains and linguistic styles. Utilizing state-of-the-art Quantized Low-Rank Adaptation (QLoRA) techniques, Cognitivess delivers high-quality text generation while maintaining exceptional efficiency.

Key features:

Built on a custom-designed architecture inspired by LLaMA, optimized for versatility and performance
Trained on a rich tapestry of data sources, including scientific literature, creative writing, multilingual corpora, and real-world conversational data
Employs advanced few-shot learning capabilities, allowing it to quickly adapt to new tasks with minimal examples
Capable of generating text in multiple languages, with particular strength in English and Romanian
Specialized in tasks such as text generation, sentiment analysis, and complex problem-solving across various domains
Incorporates ethical AI principles, with built-in safeguards against generating harmful or biased content

Cognitivess aims to serve as more than just an AI assistant; it's designed to be a knowledgeable companion capable of engaging in substantive discussions on topics ranging from cutting-edge technology to classical literature. Whether you need help with data analysis, creative storytelling, or exploring abstract concepts, Cognitivess is equipped to provide nuanced and contextually appropriate responses.

This model represents Cognitivess AI's commitment to pushing the boundaries of natural language processing. By combining vast knowledge with advanced reasoning capabilities, Cognitivess strives to bridge the gap between artificial and human intelligence, opening new possibilities for AI applications across various industries and academic fields.

Under the Cognitivess Open Model License, Cognitivess AI confirms:

Models are commercially usable.
You are free to create and distribute Derivative Models.
Cognitivess does not claim ownership to any outputs generated using the Models or Derivative Models.

Intended use

Cognitivess is a multilingual chat model designed to support a variety of languages including English, Romanian, Spanish, French, German, and many more, intended for diverse language applications.

Model Developer: Cognitivess AI

Model Dates: Cognitivess was trained between July 2024.

Data Freshness: The pretraining data has a cutoff of June 2024. Training will continue beyond the current data cutoff date to incorporate new data as it becomes available.

Model Architecture:

Cognitivess model architecture is Transformer-based and trained with a sequence length of 8192 tokens.

Architecture Type: Transformer (auto-regressive language model)

Try this model on bella.cognitivess.com now.

Usage

To use this model, first install the custom package:

# Install required packages
!pip install git+https://huggingface.co/CognitivessAI/cognitivess

Then, you can use the model like this:

import cognitivess_model  # Ensure this imports the custom model package

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "CognitivessAI/cognitivess"

# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Load the model
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float32,
    device_map="auto"
)

messages = [
    {"role": "user", "content": "Explain how large language models work in detail."},
]

input_ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)

outputs = model.generate(
    input_ids,
    do_sample=True,
    temperature=0.5,
    max_new_tokens=1024
)

response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))

Usage with LORA + Quantized Versions through bitsandbytes

To use this model, first install the custom package:

# Install required packages
!pip install git+https://huggingface.co/CognitivessAI/cognitivess
!pip install bitsandbytes
!pip install peft

Then, you can use the model like this:

import cognitivess_model  # Ensure this imports the custom model package

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel, get_peft_config, LoraConfig
import torch

model_id = "CognitivessAI/cognitivess"

# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Define the quantization configuration
quantization_config = {
    "load_in_8bit": True,
    "llm_int8_threshold": 6.0
}

# Load the model with 8-bit quantization
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float32,
    device_map="auto",
    **quantization_config
)

# Define the fine-tuning configuration
fine_tuning_config = LoraConfig(
    r=8,
    lora_alpha=16,
    lora_dropout=0.1,
    target_modules=["q_proj", "v_proj"]
)

# Apply parameter-efficient fine-tuning (PEFT) using QLoRA
model = PeftModel(model, fine_tuning_config)

# Prepare the messages
messages = [
    {"role": "user", "content": "Explain how large language models work in detail."},
]

# Tokenize the input
input_ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)

# Define the inference parameters
inference_params = {
    "max_new_tokens": 8192,
    "temperature": 0.7,
    "top_p": 0.95,
    "do_sample": True
}

# Generate the response
outputs = model.generate(
    input_ids,
    **inference_params
)

# Decode and print the response
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))

Contact: hello@cognitivess.com