show me the list of text models (text encoders) for Llava type (only that one), Except…
- User
show me the list of text models (text encoders) for Llava type (only that one), Except meta Llama network.
Assistant
LLaVA (Large Language-and-Vision Assistant) is a type of large multimodal model that typically combines a pre-trained vision encoder, such as CLIP, with a large language model (LLM) using a connection mechanism, often a simple projection layer or a multi-layer perceptron (MLP)[1][2][3]. While the original LLaVA and earlier versions like LLaVA 1.5 primarily used Vicuna (which is based on Llama) as their language model backbone, later versions and related models have incorporated other LLMs[1][2][3].
Excluding models based on Meta's Llama family (including Vicuna), here are some text encoders (language models) used in LLaVA-type models:
- Mistral-7B: LLaVA 1.6 (LLaVA-NeXT) includes variants that utilize the Mistral-7B language model[3].
- Nous-Hermes-2-Yi-34B: LLaVA 1.6 (LLaVA-NeXT) also considers this model, which is based on the Yi-34B LLM[3][8]. Yi is a large language model trained from scratch on a bilingual Chinese-English dataset[4].
- Qwen-1.5 (72B and 110B): LLaVA-NeXT has expanded to support Qwen-1.5 models with 72B and 110B parameters[8].
- Gemma: There is a LLaVA-style model variant called LLaVA-Gemma-2B, which uses the Gemma language model[6].
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.