Quantization made by Richard Erkhov.

avi-1.1b - GGUF

Model creator: https://huggingface.co/ashercn97/
Original model: https://huggingface.co/ashercn97/avi-1.1b/

Name	Quant method	Size
avi-1.1b.Q2_K.gguf	Q2_K	0.4GB
avi-1.1b.Q3_K_S.gguf	Q3_K_S	0.47GB
avi-1.1b.Q3_K.gguf	Q3_K	0.51GB
avi-1.1b.Q3_K_M.gguf	Q3_K_M	0.51GB
avi-1.1b.Q3_K_L.gguf	Q3_K_L	0.55GB
avi-1.1b.IQ4_XS.gguf	IQ4_XS	0.57GB
avi-1.1b.Q4_0.gguf	Q4_0	0.59GB
avi-1.1b.IQ4_NL.gguf	IQ4_NL	0.6GB
avi-1.1b.Q4_K_S.gguf	Q4_K_S	0.6GB
avi-1.1b.Q4_K.gguf	Q4_K	0.62GB
avi-1.1b.Q4_K_M.gguf	Q4_K_M	0.62GB
avi-1.1b.Q4_1.gguf	Q4_1	0.65GB
avi-1.1b.Q5_0.gguf	Q5_0	0.71GB
avi-1.1b.Q5_K_S.gguf	Q5_K_S	0.71GB
avi-1.1b.Q5_K.gguf	Q5_K	0.73GB
avi-1.1b.Q5_K_M.gguf	Q5_K_M	0.73GB
avi-1.1b.Q5_1.gguf	Q5_1	0.77GB
avi-1.1b.Q6_K.gguf	Q6_K	0.84GB
avi-1.1b.Q8_0.gguf	Q8_0	1.09GB

Original model description:

license: apache-2.0 language: - he - en pipeline_tag: text-generation

Avi 1.1B

Avi is a fine tune of TinyLlama. While TinyLlama originally only spoke English, Avi was trained on Hebrew data, so it speaks hebrew.

Data

Avi was trained on the Platypus and Alpaca-GPT4 datasets, machine translated into Hebrew. While I know this is not the optimal approach, it is what I did.