README.md

metadata

license: apache-2.0
language:
  - en
base_model:
  - Qwen/Qwen2.5-7B-Instruct
pipeline_tag: text-generation
library_name: transformers
tags:
  - opus
  - code
  - qwen2.5
  - cot
  - lcot

Taurus-Opus-7B-Elite

Taurus-Opus-7B-Elite is based on a 7B-parameter architecture inspired by Qwen 2.5, optimized to deliver exceptional reasoning, contextual understanding, and problem-solving capabilities. It has been fine-tuned with a focus on chain-of-thought (CoT) reasoning using a specialized dataset for tasks requiring logical deductions and multi-step problem-solving. Despite its reduced parameter count, Taurus-Opus-7B-Elite remains highly efficient and versatile, tailored for a range of applications such as instruction-following, structured data processing, and multilingual tasks.

Key Improvements

1. Compact Yet Powerful:
   Despite being a 7B-parameter model, Taurus-Opus demonstrates powerful reasoning and understanding capabilities comparable to larger models due to advanced optimization techniques.

2. Enhanced Efficiency:
   Optimized for faster inference and reduced computational costs, making it suitable for deployments on devices with limited resources.

3. Instruction Following:
   Improved capabilities in understanding and executing complex instructions while generating long texts (up to 4K tokens).

4. Structured Data Processing:
   Excels at analyzing tables, JSON, and other structured data formats, ensuring accurate and structured outputs.

5. Multilingual Proficiency:
   Supports 20+ languages, maintaining accuracy and fluency in common languages such as English, Chinese, Spanish, and French.

6. Streamlined Long-Context Support:
   Supports up to 64K tokens, providing robust contextual understanding for long-chain reasoning tasks.

Quickstart with transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Taurus-Opus-7B-Elite"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "Explain why reasoning is critical in solving complex problems."
messages = [
    {"role": "system", "content": "You are Taurus, an advanced AI assistant optimized for reasoning and problem-solving."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=256
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

Intended Use

1. Reasoning and Contextual Understanding:
   Tailored for tasks that require logical deductions and contextual analysis, suitable for educational and professional use cases.

2. Mathematical Reasoning:
   Adept at solving mathematical problems and calculations, making it ideal for STEM applications.

3. Code Assistance:
   Provides support for generating, debugging, and optimizing code in a variety of programming languages.

4. Multilingual Tasks:
   Enables global applications, including multilingual content generation, translation, and conversational AI.

5. Content Generation:
   Generates high-quality long-form text for reports, articles, and other professional documents.

Limitations

1. Reduced Parameter Count:
   While efficient, it may not achieve the same depth of understanding as larger models like 14B-parameter counterparts in some complex tasks.

2. Hardware Requirements:
   Though lighter than larger models, it still requires a GPU or high-performance CPU for optimal performance.

3. Multilingual Accuracy:
   Performance may vary for less-resourced languages, with minor inaccuracies in nuanced translations.

4. Error Propagation in Long Outputs:
   Similar to larger models, early output errors in long-text generation can affect the coherence of the final text.

5. Prompt Sensitivity:
   Requires well-structured prompts for best performance, necessitating some user familiarity with prompt design.