Ganymede-Llama-3.3-3B-Preview

Ganymede-Llama-3.3-3B-Preview is based on the Llama-3.2-3B-Instruct architecture, featuring unlocked abliterated capabilities and improved mathematical analysis. Fine-tuned on a high-quality synthetic dataset derived from Llama's Instruct series, it excels in chain-of-thought (CoT) reasoning, logical problem-solving, and structured data comprehension. The model is ideal for complex reasoning tasks, instruction-following, and text generation, with superior adaptability across multi-turn conversations and long-context tasks.

Key Improvements

1. Unlocked Abliterated Reasoning: Enhanced multi-step problem-solving, logical deduction, and contextual analysis.
2. Mathematical & Analytical Excellence: Stronger capabilities in math problem-solving, theorem proving, and complex numerical analysis.
3. Fine-Tuned Instruction Following: Generates structured responses (e.g., JSON, XML, Markdown) and long-form text (4K+ tokens).
4. Extended Long-Context Support: Handles up to 128K tokens with improved memory retention and coherence over long passages.
5. Advanced Adaptability: Excels in role-playing, multi-turn dialogues, and diverse system prompts.
6. Multilingual Proficiency: Supports over 20 languages, including English, Chinese, French, Spanish, Portuguese, German, and more.

Quickstart with Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Ganymede-Llama-3.3-3B-Preview"

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

prompt = "Explain the concept of logical reasoning in AI."
messages = [
    {"role": "system", "content": "You are an expert AI assistant specialized in reasoning, logic, and mathematical analysis."},
    {"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]
print(response)

Intended Use

• Advanced Logical & Analytical Reasoning: Designed for multi-step problem-solving, deductive reasoning, and cognitive tasks.
• Enhanced Mathematical Computation: Excels in numerical analysis, theorem proving, symbolic reasoning, and complex calculations.
• Code Generation & Debugging: Generates optimized code, detects bugs, and enhances programming workflows.
• Structured Data Processing: Handles tables, JSON, and structured formats for data-centric applications.
• Multilingual Reasoning & Translation: High proficiency across 20+ languages for global AI applications.
• Extended Text Generation: Ideal for generating technical documentation, research papers, instructional guides, and in-depth reports.

Limitations

1. Moderate Computational Requirements: Requires mid-to-high-end consumer GPUs for optimal inference.
2. Language-Specific Variability: Performance may differ across languages, particularly for low-resource languages.
3. Potential Error Accumulation: Long-form text generation may introduce inconsistencies over extended outputs.
4. Limited Real-World Awareness: Knowledge is restricted to training data and may not reflect recent world events.
5. Prompt Sensitivity: The quality of responses depends on prompt clarity and specificity.