Modelpublic

kejcao/llava-phi-2-GGUF

sourceHugging Facemitupdated 3y agoView on Hugging Face

3likes140downloads

Model Card

GGUF Quantized LLaVa Phi-2 3B

Name	Quant method	Bits	Size	Max RAM required	Use case
ggml-model-Q2_K.gguf	Q2_K	2	1.17 GB	3.67 GB	smallest, significant quality loss - not recommended for most purposes
ggml-model-Q3_K_S.gguf	Q3KS	3	1.25 GB	3.75 GB	very small, high quality loss
ggml-model-Q3_K_M.gguf	Q3KM	3	1.48 GB	3.98 GB	very small, high quality loss
ggml-model-Q4_0.gguf	Q4_0	4	1.60 GB	4.10 GB	legacy; small, very high quality loss - prefer using Q3KM
ggml-model-Q3_K_L.gguf	Q3KL	3	1.60 GB	4.10 GB	small, substantial quality loss
ggml-model-Q4_K_S.gguf	Q4KS	4	1.62 GB	4.12 GB	small, greater quality loss
ggml-model-Q4_K_M.gguf	Q4KM	4	1.79 GB	4.29 GB	medium, balanced quality - recommended
ggml-model-Q5_0.gguf	Q5_0	5	1.93 GB	4.43 GB	legacy; medium, balanced quality - prefer using Q4KM
ggml-model-Q5_K_S.gguf	Q5KS	5	1.93 GB	4.43 GB	large, low quality loss - recommended
ggml-model-Q5_K_M.gguf	Q5KM	5	2.07 GB	4.57 GB	large, very low quality loss - recommended
ggml-model-Q6_K.gguf	Q6_K	6	2.29 GB	4.79 GB	very large, extremely low quality loss
ggml-model-Q8_0.gguf	Q8_0	8	2.96 GB	5.46 GB	very large, extremely low quality loss - not recommended

—Developed by: LAION, SkunkworksAI & Ontocord
—Model type: LLaVA is an open-source chatbot trained by fine-tuning Phi-2 on GPT-generated multimodal instruction-following data. It is an auto-regressive language model, based on the transformer architecture
—Finetuned from model: Phi-2
—License: MIT
—Demo: llava-phi-2-3b-demo

Model	Parameters	SQA	GQA	TextVQA	POPE
LLaVA-1.5	7.3B	68.0	62.0	58.3	85.3
MC-LLaVA-3B	3B	-	49.6	38.59	-
LLaVA-Phi	3B	68.4	-	48.6	85.0
moondream1	1.6B	-	56.3	39.8	-
llava-phi-2-3b	3B	69.0	51.2	47.0	86.0

Model	BLEU_1	BLEU_2	BLEU_3	BLEU_4	METEOR	ROUGE_L	CIDEr	SPICE
llava-1.5-7b	75.8	59.8	45	33.3	29.4	57.7	108.8	23.5
llava-phi-2-3b	67.7	50.5	35.7	24.2	27.0	52.4	85.0	20.7