# Multi-round dialogue optimization

FlashVL is a new approach to optimizing vision-language models (VLMs) for real-time applications, aiming to achieve ultra-low latency and high throughput without sacrificing accuracy.

A large language model with 4B parameters based on the Hugging Face transformers library, supporting functions such as text generation, thinking mode switching, tool invocation, and long text processing.

This is an MLX-format text generation model converted from Qwen/Qwen3-0.6B, supporting Chinese and English text generation tasks.

Qwen3-0.6B-8bit is an 8-bit quantized version converted from Qwen/Qwen3-0.6B, a text generation model suitable for the MLX framework.

A quantized version based on Google Gemma 3's 27-billion parameter instruction-tuned model, generated using quantization-aware training (QAT) weights, supporting multiple quantization levels to meet different hardware requirements.

Gemma 2 27B IT is a 4-bit large language model based on AutoAWQ quantization, suitable for dialogue and instruction-following tasks.

This is a lightweight version of Llama-4-Scout-17B-16E-Instruct, providing users with a more streamlined usage option.

xLAM-2 is a large action model built on an advanced data synthesis and training pipeline. It excels in multi-round dialogue and tool usage, and can transform user intentions into executable actions.

Gemma 3.4B IT is a lightweight open-source large language model released by Google. Based on a parameter scale of 3.4B, it is suitable for dialogue and instruction following tasks.

Gemma-3-4b-it is a lightweight language model released by Google, based on the Gemma architecture and suitable for text generation tasks.

Llama 3.1 Swallow is a series of large language models built on Meta Llama 3.1. It enhances Japanese language capabilities through continuous pre-training while retaining English language capabilities.

Llama 3.1 Swallow is a series of large language models built on Meta Llama 3.1. It enhances Japanese capabilities through continuous pre-training while retaining English capabilities.

Lumimaid 0.2 is a model based on Meta-Llama-3.1-70B-Instruct. Compared with version 0.1, there has been a huge improvement in the dataset. After data cleaning and optimization, it provides a better user experience.

Eliminate the rejection response problem of the Llama-3.1-8B-Instruct model through LoRA technology

Lumimaid 0.2 is a model optimized based on Meta-Llama-3.1-8B-Instruct. Its performance has been significantly improved through data cleaning and expansion, providing higher-quality text generation services.

Mistral-Nemo-Instruct-2407 is a large language model fine-tuned for instructions based on the Mistral architecture, suitable for various natural language processing tasks.

Hermes-2 Θ Llama-3 8B is a powerful model that combines the advantages of Hermes 2 Pro and Meta's Llama-3 Instruct, and performs well in various tasks. It supports multiple prompt formats and function calls.

KARAKURI LM is a pre-trained language model built on Llama 2, which enhances Japanese processing capabilities and is further pre-trained on Japanese and multilingual corpora.

The first open commercial-use German base language model built on Llama-2, focusing on German language processing