Tanuki-8B-dpo-v1.0
| Property | Value |
|---|---|
| Parameter Count | 7.51B |
| License | Apache 2.0 |
| Languages | Japanese, English |
| Training Data | 1.3T tokens |
| Model Type | Transformer-based LLM |
What is Tanuki-8B-dpo-v1.0?
Tanuki-8B-dpo-v1.0 is a sophisticated bilingual language model developed by the GENIAC project at the University of Tokyo. This model represents a significant achievement in Japanese-English language modeling, having undergone extensive pre-training with 1.3T tokens and refined through DPO (Direct Preference Optimization) for enhanced conversational capabilities.
Implementation Details
The model utilizes a transformer architecture and supports BF16 precision. It's been specifically optimized for both Japanese and English language processing, with various quantized versions available including AWQ 4-bit, GPTQ 4-bit/8-bit, and GGUF formats.
- Follows Japanese Alpaca prompt format
- Supports both single-turn and multi-turn conversations
- Implements chat template functionality
- Offers streaming capabilities for real-time response generation
Core Capabilities
- Strong performance in Japanese MT-Bench evaluation (7.24 average score)
- Excels in humanities (9.1), STEM (9.35), and writing (9.05) tasks
- Efficient handling of bilingual conversations
- Specialized in instruction-following scenarios
Frequently Asked Questions
Q: What makes this model unique?
The model stands out for its specialized Japanese-English capabilities and extensive pre-training on 1.3T tokens, making it particularly effective for bilingual applications. Its DPO fine-tuning ensures high-quality conversational abilities.
Q: What are the recommended use cases?
The model is ideal for Japanese-English conversational AI applications, academic discussion, writing assistance, and general knowledge tasks. It shows particularly strong performance in humanities, STEM, and writing-related queries.
