shisa-ai/ablation-54-nemojp-shisa-v2-mistral-nemo-japanese-12b

TEXT GENERATIONPricing:Input $0.87 / Cached $0.2 / Output $0.99Concurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Mar 14, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The shisa-ai/ablation-54-nemojp-shisa-v2-mistral-nemo-japanese-12b is a 12 billion parameter language model fine-tuned by shisa-ai, based on cyberagent/Mistral-Nemo-Japanese-Instruct-2408. It was trained on the shisa-ai/shisa-v1-athenev2-reannotated-filtered and shisa-ai/shisa-v2-roleplaying datasets, featuring a context length of 32768 tokens. This model is optimized for Japanese language tasks, particularly those involving instruction following and role-playing scenarios.

Loading preview...

Model Overview

This model, ablation-54-nemojp-shisa-v2-mistral-nemo-japanese-12b, is a 12 billion parameter language model developed by shisa-ai. It is a fine-tuned variant of the cyberagent/Mistral-Nemo-Japanese-Instruct-2408 base model, specifically enhanced for Japanese language processing. The model was trained using Axolotl, incorporating Liger plugins for improved performance, and supports a substantial context length of 32768 tokens.

Key Capabilities

  • Japanese Language Proficiency: Fine-tuned on specialized Japanese datasets, making it suitable for nuanced Japanese text generation and understanding.
  • Instruction Following: Enhanced through training on instruction-based datasets, enabling it to respond effectively to user prompts.
  • Role-Playing Scenarios: Specifically trained on the shisa-ai/shisa-v2-roleplaying dataset, indicating a strong capability for engaging in conversational and character-based interactions.
  • Efficient Training: Utilizes advanced training techniques like flash_attention and gradient_checkpointing for optimized resource usage.

Good for

  • Japanese Chatbots: Ideal for developing conversational AI agents that require strong Japanese language skills and the ability to maintain character or follow specific instructions.
  • Content Generation: Suitable for generating Japanese text in various styles, especially for interactive or narrative purposes.
  • Research and Development: Provides a robust base for further experimentation and fine-tuning on specific Japanese NLP tasks, leveraging its specialized training data.

Finetunes
1 models