You can deploy this model and start earning money today!

RichardErkhov/
Alepach_-_notHumpback-Myx-gguf

Name: Alepach_-_notHumpback-Myx-gguf
Availability: InStock

Views

No views yet

Quantization made by Richard Erkhov.

Github

Discord

Request more models

notHumpback-Myx - GGUF

Model creator: https://huggingface.co/Alepach/
Original model: https://huggingface.co/Alepach/notHumpback-Myx/

Name	Quant method	Size
notHumpback-Myx.Q2_K.gguf	Q2_K	1.27GB
notHumpback-Myx.IQ3_XS.gguf	IQ3_XS	1.38GB
notHumpback-Myx.IQ3_S.gguf	IQ3_S	1.44GB
notHumpback-Myx.Q3_K_S.gguf	Q3_K_S	1.44GB
notHumpback-Myx.IQ3_M.gguf	IQ3_M	1.49GB
notHumpback-Myx.Q3_K.gguf	Q3_K	1.57GB
notHumpback-Myx.Q3_K_M.gguf	Q3_K_M	1.57GB
notHumpback-Myx.Q3_K_L.gguf	Q3_K_L	1.69GB
notHumpback-Myx.IQ4_XS.gguf	IQ4_XS	1.71GB
notHumpback-Myx.Q4_0.gguf	Q4_0	1.79GB
notHumpback-Myx.IQ4_NL.gguf	IQ4_NL	1.79GB
notHumpback-Myx.Q4_K_S.gguf	Q4_K_S	1.8GB
notHumpback-Myx.Q4_K.gguf	Q4_K	1.88GB
notHumpback-Myx.Q4_K_M.gguf	Q4_K_M	1.88GB
notHumpback-Myx.Q4_1.gguf	Q4_1	1.95GB
notHumpback-Myx.Q5_0.gguf	Q5_0	2.11GB
notHumpback-Myx.Q5_K_S.gguf	Q5_K_S	2.11GB
notHumpback-Myx.Q5_K.gguf	Q5_K	2.16GB
notHumpback-Myx.Q5_K_M.gguf	Q5_K_M	2.16GB
notHumpback-Myx.Q5_1.gguf	Q5_1	2.28GB
notHumpback-Myx.Q6_K.gguf	Q6_K	2.46GB
notHumpback-Myx.Q8_0.gguf	Q8_0	3.19GB

Original model description:

base_model: meta-llama/Llama-3.2-3B library_name: transformers model_name: notHumpback-Myx tags:

generated_from_trainer
trl
sft license: apache-2.0 datasets:
OpenAssistant/oasst1

notHumpback-Myx

This model follows the Humpback architecture, proposed in the paper Self-Alignment with Instruction Backtranslation by Li et al.

It represents the "backward model", which is used to generate instructions from web texts. These are considered as possible model outputs.

Humpback uses instruction backtranslation on a web corpus to generate input-output pairs (self-augmentation), creating a richer dataset for fine-tuning models without the need for additional manual annotation. The model then iteratively curates the created dataset, scoring the pairs by quality, and is then finetuned on the resulting subset of all pairs with the highest possible score (self-curation).

Varying from the original paper, this model is a fine-tuned version of meta-llama/Llama-3.2-3B. It has been trained using TRL.

The dataset used to train this model has been sampled from the oasst1 dataset. In order to achieve the "backward" structure, the model is trained on output-input pairs.

Framework versions

TRL: 0.12.1
Transformers: 4.46.3
Pytorch: 2.5.1
Datasets: 3.1.0
Tokenizers: 0.20.3

Citations

Original paper:

bibtex

@misc{li2023selfalignment,
    title={Self-Alignment with Instruction Backtranslation},
    author={Xian Li and Ping Yu and Chunting Zhou and Timo Schick and Luke Zettlemoyer and Omer Levy and Jason Weston and Mike Lewis},
    year={2023},
    eprint={2308.06259},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

Cite TRL as: