CoolFace
Modelpublic

juyoung-trl/Trillion-7B-preview-AWQ

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face
0likes33downloads
Model Card

Trillion-7B-preview

<p align="center"> <picture> <source media="(prefers-color-scheme: dark)" srcset="assets/SignitureTrillionWhiteBGresized.jpg", width="300", style="margin: 40 auto;"> <img src="assets/SignitureTrillionBlackBGresized.jpg" alt="logo", width="300", style="margin: 40 auto;"> </picture>

Introduction

We introduce Trillion-7B-preview, a preview of our latest large language model designed to push the boundaries of multilingual scalability and performance.

When comparing performance to training FLOPs for Trillion-7B-preview with competitive models, our model pushes the Pareto frontier, achieving around 66.5% average performance while using significantly fewer compute (~9.3×10²² FLOPs). It outperforms models like Mistral-7B-Instruct-v0.3 and SOLAR-10.7B-Instruct-v1.0 while remaining competitive with models requiring 3-8× more compute such as Qwen2.5-7B-Instruct and EXAONE-3.5-7.8B-Instruct. For full benchmark results, see tables below.

<p align="center"> <img src="assets/frontier.png" alt="Average Performance vs. Approximate Training FLOPs" width="700"> </p>

  • —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: 4,096
  • —Number of Tokens seen: 2T
  • —Vocab Size: 128,128

Quickstart

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

python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "trillionlabs/Trillion-7B-preview"

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

prompt = "Tell me a hilarious knock knock joke."
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["input_ids"],
    attention_mask=model_inputs["attention_mask"],
    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)

"""
Sure! Here's a classic knock-knock joke that's guaranteed to make you chuckle:
Knock, knock.
Who's there?
Lettuce.
Lettuce who?
Lettuce in, it's too cold out here!
"""

Evaluation

We select a wide variety of benchmarks that evaluate general reasoning, knowledge recall, coding abilities, mathematical reasoning, and instruction following capabilities. We evaluated Trillion-7B-preview along with several leading large language models of similar size. Our model especially demonstrates strong performance on Korean benchmarks.

<details> <summary> Full evaluation settings </summary>

BenchmarkLanguageEvaluation SettingMetric
General Reasoning and Reading Comprehension
• HellaSwagEnglish0-shotaccuracy
• TruthfulQA_mc1English6-shotaccuracy
• TruthfulQA_mc2English6-shotaccuracy
• ARC:CEnglish0-shotaccuracy
• HAERAEKorean3-shotaccuracy
• KoBESTKorean5-shotaccuracy
• BBHEnglish0-shot, CoTaccuracy
• xwinograd_enEnglish0-shotaccuracy
• xwinograd_jpJapanese0-shotaccuracy
• xwinograd_zhChinese0-shotaccuracy
Knowledge Recall
• KMMLUKorean5-shotaccuracy
• MMLUEnglish5-shotaccuracy
• Global-MMLU-Lite-enEnglish5-shotaccuracy
• Global-MMLU-Lite-koKorean5-shotaccuracy
• Global-MMLU-Lite-jaJapanese5-shotaccuracy
• Global-MMLU-Lite-zhChinese5-shotaccuracy
Coding
• HumanEvalEnglish0-shot, CoTpass@1
• MBPPEnglish0-shot, CoTpass@1
Mathematical Reasoning
• GSM8kEnglish0-shot, CoTexact-match
• MATHEnglish0-shot, CoTexact-match
• GPQAEnglish4-shotaccuracy
• HRM8kKorean0-shot, CoTexact-match
Instruction Following and Chat
• IFEvalEnglish0-shotstrict-average
• koIFEval*Korean0-shotstrict-average
• MT-Bench**EnglishLLM-as-a-judge (gpt-4o-2024-08-06)LLM score
• KO-MT-Bench**KoreanLLM-as-a-judge (gpt-4o-2024-08-06)LLM score
• LogicKor**KoreanLLM-as-a-judge (gpt-4o-2024-08-06)LLM score
  • —*Note that koIFEval is our in-house evaluation benchmark for assessing instruction-following capabilities in Korean.
  • —**Note that MT-Bench, KO-MT-Bench, and LogicKor use a 10-point scale.

</details>

Benchmark Results

General Reasoning and Factuality

BenchmarkTrillion-7B-previewEXAONE-3.5-7.8B-Instructgemma-2-9b-itLlama-3.1-8B-InstructQwen2.5-7B-InstructSOLAR-10.7B-Instruct-v1.0Mistral-7B-Instruct-v0.3
HellaSwag58.9460.0459.7259.8161.9768.7265.79
TruthfulQA_mc136.1040.6442.9638.0747.7456.1842.47
TruthfulQA_mc254.1059.7460.0954.5464.7270.6459.41
ARC:C54.4456.4062.9753.5852.9960.0758.11
HAERAE80.0276.0868.0163.1565.1760.8647.75
KoBEST79.6178.5779.9870.0979.2475.2066.50
KMMLU48.0945.3946.6641.4150.1541.6633.59
MMLU63.5265.6572.2468.3274.2365.2061.84
Global-MMLU-Lite-en67.7569.5076.2567.5077.2571.7565.50
Global-MMLU-Lite-ko60.7560.0064.2554.0059.2553.7543.00
Global-MMLU-Lite-ja60.7545.7566.5054.5065.7550.7550.00
Global-MMLU-Lite-zh59.5050.0063.7560.2568.7557.0047.25
BBH41.9453.3028.7743.1653.6852.9145.09
xwinograd_en87.7887.1089.5588.0985.6387.3588.39
xwinograd_jp79.9874.4580.9276.0272.8972.5870.70
xwinograd_zh73.8169.4468.0676.1981.5574.6071.83

Coding

BenchmarkTrillion-7B-previewEXAONE-3.5-7.8B-Instructgemma-2-9b-itLlama-3.1-8B-InstructQwen2.5-7B-InstructSOLAR-10.7B-Instruct-v1.0Mistral-7B-Instruct-v0.3
HumanEval55.4879.2660.9867.6881.7134.7636.59
MBPP40.4061.408.4039.2051.0029.4036.00

Mathematical Reasoning

BenchmarkTrillion-7B-previewEXAONE-3.5-7.8B-Instructgemma-2-9b-itLlama-3.1-8B-InstructQwen2.5-7B-InstructSOLAR-10.7B-Instruct-v1.0Mistral-7B-Instruct-v0.3
GSM8k72.2587.7973.6974.9888.8662.9335.94
MATH32.7070.68-38.3071.5014.3812.12
GPQA32.8138.6136.8330.5834.1528.3532.59
HRM8k30.1038.9916.04-41.5120.687.89

Instruction Following and Chat

BenchmarkTrillion-7B-previewEXAONE-3.5-7.8B-Instructgemma-2-9b-itLlama-3.1-8B-InstructQwen2.5-7B-InstructSOLAR-10.7B-Instruct-v1.0Mistral-7B-Instruct-v0.3
IFEval79.1381.4275.4874.9375.8551.6152.64
koIFEval66.5854.6543.3036.0748.5526.1234.22
MT-Bench6.536.75-6.327.866.766.84
KO-MT-Bench6.216.70-4.276.475.574.59
LogicKor8.149.258.336.457.991.854.76

Limitations

  • —Language Support: The model is optimized for English, Korean, Japanese, and Chinese. Usage with other languages may result in degraded performance.
  • —Knowledge Cutoff: The model's information is limited to data available up to August 2023.
  • —Safety Mechanisms: This release does not yet include comprehensive safety features. Future updates will address this area.
  • —Release Status: This is a preliminary release version with planned enhancements and updates forthcoming.

License

This model repository is licensed under the Apache-2.0 License.

Citation

@article{trillion7Bpreview,
  title={Trillion-7B-preview},
  author={trillionlabs},
  year={2025},
  url={https://huggingface.co/trillionlabs/Trillion-7B-preview}
}

Contact

For inquiries, please contact: info@trillionlabs.co