Tri-7B

Introduction

We introduce Tri-7B, the next generation model following Trillion-7B-preview, that continues to push the boundaries of efficient training while achieving exceptional performance at the 7B parameter scale.

Key Highlights

• Enhanced Reasoning: Modified training dataset mixture specifically optimized for reasoning capabilities
• Advanced Post-Training: Significantly improved RL training pipeline focusing on mathematical reasoning and everyday usage
• Extended Context: Supports up to 32K context length for long-form understanding
• Multi-lingual: Specially optimized for Korean, English, and Japanese.

Our Tri-7B model represents a significant advancement over Trillion-7B-preview, achieving substantial performance improvements across all evaluated domains while maintaining the same efficient parameter count.

Model Specifications

Tri-7B

• Type: Causal Language Model
• Training Stage: Pre-training & Post-training
• Architecture: Transformer Decoder with RoPE, SwiGLU, RMSNorm
• Number of Parameters: 7.76B
• Number of Layers: 32
• Number of Attention Heads: 32
• Context Length: 32,768
• Vocab Size: 128,256

Quickstart

Here is a code snippet with apply_chat_template that demonstrates how to load the tokenizer and model and generate text.

Tri-7B Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "trillionlabs/Tri-7B"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "Explain the concept of quantum computing in simple terms."
messages = [
    {"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=512
)
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)

Tri-7B is also available with vLLM and SGLang!

# vLLM
vllm serve trillionlabs/Tri-7B --dtype bfloat16 --max-model-len 32768

# vLLM with custom options
vllm serve trillionlabs/Tri-7B \
    --dtype bfloat16 \
    --max-model-len 32768 \
    --gpu-memory-utilization 0.95 \
    --port 8000

# SGLang
python3 -m sglang.launch_server --model-path trillionlabs/Tri-7B --dtype bfloat16

# SGLang with custom options
python3 -m sglang.launch_server \
    --model-path trillionlabs/Tri-7B \
    --dtype bfloat16 \
    --context-length 32768 \
    --port 30000 \
    --host 0.0.0.0

Evaluation

We evaluated Tri-7B across a comprehensive suite of benchmarks assessing general reasoning, knowledge recall, coding abilities, mathematical reasoning, and instruction-following capabilities. Compared to our previous generation model Trillion-7B-preview, Tri-7B achieves significant gains across all domains.

• *Note that koIFEval, systemIFEval, and KoRuler are our in-house evaluation benchmarks adapted for Korean to better assess model capabilities in Korean language tasks.
• **Note that MT-Bench, KO-MT-Bench, and LogicKor use a 10-point scale.

Benchmark Results

Models compared:

• Tri-7B (Next Generation)
• Trillion-7B-preview (Previous Generation)

General Reasoning and Factuality

Coding

Mathematical Reasoning

Instruction Following and Chat

Limitations

• Language Support: The model is optimized for English, Korean, and Japanese. Usage with other languages may result in degraded performance.
• Knowledge Cutoff: The model's information is limited to data available up to Febuary, 2025.

License

This model is licensed under the Apache License 2.0.

Contact

For inquiries, please contact: info@trillionlabs.co