You can deploy this model and start earning money today!

RichardErkhov/
arnavgrg_-_mistral-7b-instruct-nf4-fp16-upscaled-gguf

Name: arnavgrg_-_mistral-7b-instruct-nf4-fp16-upscaled-gguf
Availability: InStock

Views

No views yet

Quantization made by Richard Erkhov.

Github

Discord

Request more models

mistral-7b-instruct-nf4-fp16-upscaled - GGUF

Model creator: https://huggingface.co/arnavgrg/
Original model: https://huggingface.co/arnavgrg/mistral-7b-instruct-nf4-fp16-upscaled/

Name	Quant method	Size
mistral-7b-instruct-nf4-fp16-upscaled.Q2_K.gguf	Q2_K	2.53GB
mistral-7b-instruct-nf4-fp16-upscaled.IQ3_XS.gguf	IQ3_XS	2.81GB
mistral-7b-instruct-nf4-fp16-upscaled.IQ3_S.gguf	IQ3_S	2.96GB
mistral-7b-instruct-nf4-fp16-upscaled.Q3_K_S.gguf	Q3_K_S	2.95GB
mistral-7b-instruct-nf4-fp16-upscaled.IQ3_M.gguf	IQ3_M	3.06GB
mistral-7b-instruct-nf4-fp16-upscaled.Q3_K.gguf	Q3_K	3.28GB
mistral-7b-instruct-nf4-fp16-upscaled.Q3_K_M.gguf	Q3_K_M	3.28GB
mistral-7b-instruct-nf4-fp16-upscaled.Q3_K_L.gguf	Q3_K_L	3.56GB
mistral-7b-instruct-nf4-fp16-upscaled.IQ4_XS.gguf	IQ4_XS	3.67GB
mistral-7b-instruct-nf4-fp16-upscaled.Q4_0.gguf	Q4_0	3.83GB
mistral-7b-instruct-nf4-fp16-upscaled.IQ4_NL.gguf	IQ4_NL	3.87GB
mistral-7b-instruct-nf4-fp16-upscaled.Q4_K_S.gguf	Q4_K_S	3.86GB
mistral-7b-instruct-nf4-fp16-upscaled.Q4_K.gguf	Q4_K	4.07GB
mistral-7b-instruct-nf4-fp16-upscaled.Q4_K_M.gguf	Q4_K_M	4.07GB
mistral-7b-instruct-nf4-fp16-upscaled.Q4_1.gguf	Q4_1	4.24GB
mistral-7b-instruct-nf4-fp16-upscaled.Q5_0.gguf	Q5_0	4.65GB
mistral-7b-instruct-nf4-fp16-upscaled.Q5_K_S.gguf	Q5_K_S	4.65GB
mistral-7b-instruct-nf4-fp16-upscaled.Q5_K.gguf	Q5_K	4.78GB
mistral-7b-instruct-nf4-fp16-upscaled.Q5_K_M.gguf	Q5_K_M	4.78GB
mistral-7b-instruct-nf4-fp16-upscaled.Q5_1.gguf	Q5_1	5.07GB
mistral-7b-instruct-nf4-fp16-upscaled.Q6_K.gguf	Q6_K	5.53GB
mistral-7b-instruct-nf4-fp16-upscaled.Q8_0.gguf	Q8_0	7.17GB

Original model description:

license: apache-2.0 tags:

text-generation-inference

This is an upscaled fp16 variant of the original Mistral-7B-Instruct-v0.1 base model by Mistral after it has been loaded with nf4 4-bit quantization via bitsandbytes. The main idea here is to upscale the linear4bit layers to fp16 so that the quantization/dequantization cost doesn't have to paid for each forward pass at inference time.

Note: The quantization operation to nf4 is not lossless, so the model weights for the linear layers are lossy, which means that this model will not work as well as the official base model.

To use this model, you can just load it via transformers in fp16:

python

import torch
from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
  "arnavgrg/mistral-7b-instruct-nf4-fp16-upscaled",
  device_map="auto",
  torch_dtype=torch.float16,
)

RichardErkhov/arnavgrg_-_mistral-7b-instruct-nf4-fp16-upscaled-gguf

Original model description:

RichardErkhov/
arnavgrg_-_mistral-7b-instruct-nf4-fp16-upscaled-gguf