RichardErkhov/Ba2han_-_TinyOpenHermes-1.1B-4k-gguf

Quantization made by Richard Erkhov.

Github

Discord

Request more models

TinyOpenHermes-1.1B-4k - GGUF

• Model creator: https://huggingface.co/Ba2han/
• Original model: https://huggingface.co/Ba2han/TinyOpenHermes-1.1B-4k/

Name	Quant method	Size
TinyOpenHermes-1.1B-4k.Q2_K.gguf	Q2_K	0.4GB
TinyOpenHermes-1.1B-4k.IQ3_XS.gguf	IQ3_XS	0.44GB
TinyOpenHermes-1.1B-4k.IQ3_S.gguf	IQ3_S	0.47GB
TinyOpenHermes-1.1B-4k.Q3_K_S.gguf	Q3_K_S	0.47GB
TinyOpenHermes-1.1B-4k.IQ3_M.gguf	IQ3_M	0.48GB
TinyOpenHermes-1.1B-4k.Q3_K.gguf	Q3_K	0.51GB
TinyOpenHermes-1.1B-4k.Q3_K_M.gguf	Q3_K_M	0.51GB
TinyOpenHermes-1.1B-4k.Q3_K_L.gguf	Q3_K_L	0.55GB
TinyOpenHermes-1.1B-4k.IQ4_XS.gguf	IQ4_XS	0.57GB
TinyOpenHermes-1.1B-4k.Q4_0.gguf	Q4_0	0.59GB
TinyOpenHermes-1.1B-4k.IQ4_NL.gguf	IQ4_NL	0.6GB
TinyOpenHermes-1.1B-4k.Q4_K_S.gguf	Q4_K_S	0.6GB
TinyOpenHermes-1.1B-4k.Q4_K.gguf	Q4_K	0.62GB
TinyOpenHermes-1.1B-4k.Q4_K_M.gguf	Q4_K_M	0.62GB
TinyOpenHermes-1.1B-4k.Q4_1.gguf	Q4_1	0.65GB
TinyOpenHermes-1.1B-4k.Q5_0.gguf	Q5_0	0.71GB
TinyOpenHermes-1.1B-4k.Q5_K_S.gguf	Q5_K_S	0.71GB
TinyOpenHermes-1.1B-4k.Q5_K.gguf	Q5_K	0.73GB
TinyOpenHermes-1.1B-4k.Q5_K_M.gguf	Q5_K_M	0.73GB
TinyOpenHermes-1.1B-4k.Q5_1.gguf	Q5_1	0.77GB
TinyOpenHermes-1.1B-4k.Q6_K.gguf	Q6_K	0.84GB
TinyOpenHermes-1.1B-4k.Q8_0.gguf	Q8_0	1.09GB

Original model description:

license: cc-by-nc-4.0 datasets: - teknium/openhermes

Task	Version	Metric	Value		Stderr
hendrycksTest-logical_fallacies	1	acc	0.3067	±	0.0362
		acc_norm	0.3067	±	0.0362
hendrycksTest-global_facts	1	acc	0.3	±	0.0461
		acc_norm	0.3	±	0.0461
hendrycksTest-abstract_algebra	1	acc	0.2700	±	0.0446
		acc_norm	0.2700	±	0.0446
hendrycksTest-college_chemistry	1	acc	0.3100	±	0.0465
		acc_norm	0.3100	±	0.0465
hendrycksTest-college_physics	1	acc	0.2157	±	0.0409
		acc_norm	0.2157	±	0.0409
hendrycksTest-formal_logic	1	acc	0.2857	±	0.0404
		acc_norm	0.2857	±	0.0404

Compared to TinyLlama-1.1B-Chat-v1.0:

Algebra UP 17.4%

Formal Logic UP 24.2%

Logical Fallacies UP 35.4%

Template Format: Alpaca

It took 4 hours to train in 1 epoch with an RTX 3090.