metadata
tags:
- quantized
- 4-bit
- AWQ
- transformers
- safetensors
- mixtral
- text-generation
- moe
- fr
- it
- de
- es
- en
- license:apache-2.0
- autotrain_compatible
- endpoints_compatible
- text-generation-inference
- region:us
model_name: Mixtral-8x22B-v0.1-AWQ
base_model: v2ray/Mixtral-8x22B-v0.1
inference: false
model_creator: v2ray
pipeline_tag: text-generation
quantized_by: MaziyarPanahi
language:
- en
- es
- de
- it
- fr
Mixtral-8x22B-v0.1-AWQ
On April 10th, @MistralAI released a model named "Mixtral 8x22B," an 176B MoE via magnet link (torrent):
- 176B MoE with ~40B active
- Context length of 65k tokens
- The base model can be fine-tuned
- Requires ~260GB VRAM in fp16, 73GB in int4
- Licensed under Apache 2.0, according to their Discord
- Available on @huggingface (community)
- Utilizes a tokenizer similar to previous models
MaziyarPanahi/Mixtral-8x22B-v0.1-AWQ is a quantized (AWQ) version of v2ray/Mixtral-8x22B-v0.1
How to use
Install the necessary packages
pip install --upgrade accelerate autoawq transformers
Example Python code
from transformers import AutoTokenizer, AutoModelForCausalLM
model_id = "MaziyarPanahi/Mixtral-8x22B-v0.1-AWQ"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id).to(0)
text = "Hello can you provide me with top-3 cool places to visit in Paris?"
inputs = tokenizer(text, return_tensors="pt").to(0)
out = model.generate(**inputs, max_new_tokens=300)
print(tokenizer.decode(out[0], skip_special_tokens=True))
Credit
MistralAI for opening the weights
v2ray for downloading, converting, and sharing it with the community Mixtral-8x22B-v0.1
philschmid for the photo he shared on his Twitter
βββββ ββββββββββββββββββ βββββββββββββββββββββββββββββββ ββββββββββββββββββββββββββββββββββ ββββββββββββββ ββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββββββ ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ βββ ββββββββββββββββββββββββββββββββ ββββββββββββββββββ ββββββββββββββββββββββββββββ βββββββββββββββββ βββββ