Quantization made by Richard Erkhov.

Yarn-Mistral-7b-128k-sharded - GGUF

Model creator: https://huggingface.co/yanismiraoui/
Original model: https://huggingface.co/yanismiraoui/Yarn-Mistral-7b-128k-sharded/

Name	Quant method	Size
Yarn-Mistral-7b-128k-sharded.Q2_K.gguf	Q2_K	2.53GB
Yarn-Mistral-7b-128k-sharded.IQ3_XS.gguf	IQ3_XS	0.34GB
Yarn-Mistral-7b-128k-sharded.IQ3_S.gguf	IQ3_S	0.05GB
Yarn-Mistral-7b-128k-sharded.Q3_K_S.gguf	Q3_K_S	0.03GB
Yarn-Mistral-7b-128k-sharded.IQ3_M.gguf	IQ3_M	0.0GB
Yarn-Mistral-7b-128k-sharded.Q3_K.gguf	Q3_K	0.0GB
Yarn-Mistral-7b-128k-sharded.Q3_K_M.gguf	Q3_K_M	0.0GB
Yarn-Mistral-7b-128k-sharded.Q3_K_L.gguf	Q3_K_L	0.0GB
Yarn-Mistral-7b-128k-sharded.IQ4_XS.gguf	IQ4_XS	0.0GB
Yarn-Mistral-7b-128k-sharded.Q4_0.gguf	Q4_0	0.0GB
Yarn-Mistral-7b-128k-sharded.IQ4_NL.gguf	IQ4_NL	0.0GB
Yarn-Mistral-7b-128k-sharded.Q4_K_S.gguf	Q4_K_S	0.0GB
Yarn-Mistral-7b-128k-sharded.Q4_K.gguf	Q4_K	0.0GB
Yarn-Mistral-7b-128k-sharded.Q4_K_M.gguf	Q4_K_M	0.0GB
Yarn-Mistral-7b-128k-sharded.Q4_1.gguf	Q4_1	0.0GB
Yarn-Mistral-7b-128k-sharded.Q5_0.gguf	Q5_0	0.0GB
Yarn-Mistral-7b-128k-sharded.Q5_K_S.gguf	Q5_K_S	0.0GB
Yarn-Mistral-7b-128k-sharded.Q5_K.gguf	Q5_K	0.0GB
Yarn-Mistral-7b-128k-sharded.Q5_K_M.gguf	Q5_K_M	0.0GB
Yarn-Mistral-7b-128k-sharded.Q5_1.gguf	Q5_1	0.0GB
Yarn-Mistral-7b-128k-sharded.Q6_K.gguf	Q6_K	0.0GB
Yarn-Mistral-7b-128k-sharded.Q8_0.gguf	Q8_0	0.0GB

Original model description:

datasets: - emozilla/yarn-train-tokenized-16k-mistral metrics: - perplexity library_name: transformers license: apache-2.0 language: - en

This repo contains a SHARDED version of: https://huggingface.co/NousResearch/Yarn-Mistral-7b-128k

Huge thanks to the publishers for their amazing work, all credits go to them: https://huggingface.co/NousResearch

Model Card: Nous-Yarn-Mistral-7b-128k

Preprint (arXiv)
GitHub

Model Description

Nous-Yarn-Mistral-7b-128k is a state-of-the-art language model for long context, further pretrained on long context data for 1500 steps using the YaRN extension method. It is an extension of Mistral-7B-v0.1 and supports a 128k token context window.

To use, pass trust_remote_code=True when loading the model, for example

model = AutoModelForCausalLM.from_pretrained("NousResearch/Yarn-Mistral-7b-128k",
  use_flash_attention_2=True,
  torch_dtype=torch.bfloat16,
  device_map="auto",
  trust_remote_code=True)

In addition you will need to use the latest version of transformers (until 4.35 comes out)

pip install git+https://github.com/huggingface/transformers

Benchmarks

Long context benchmarks:

Model	Context Window	8k PPL	16k PPL	32k PPL	64k PPL	128k PPL
Mistral-7B-v0.1	8k	2.96	-	-	-	-
Yarn-Mistral-7b-64k	64k	3.04	2.65	2.44	2.20	-
Yarn-Mistral-7b-128k	128k	3.08	2.68	2.47	2.24	2.19

Short context benchmarks showing that quality degradation is minimal:

Model	Context Window	ARC-c	Hellaswag	MMLU	Truthful QA
Mistral-7B-v0.1	8k	59.98	83.31	64.16	42.15
Yarn-Mistral-7b-64k	64k	59.38	81.21	61.32	42.50
Yarn-Mistral-7b-128k	128k	58.87	80.58	60.64	42.46

Collaborators

bloc97: Methods, paper and evals
@theemozilla: Methods, paper, model training, and evals
@EnricoShippole: Model training
honglu2875: Paper and evals

The authors would like to thank LAION AI for their support of compute for this model. It was trained on the JUWELS supercomputer.