Metrics

PPL	arc_easy	arc_challenge	piqa	winogrande	hellaswag	mmlu	QA Avg
914841.62	25.25 ± 0.89	20.14 ± 1.17	53.70 ± 1.16	48.70 ± 1.40	25.59 ± 0.44	-	34.68

Training method based on BitDistiller Paper

Safetensors

Model size

4B params

Tensor type

BF16

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Heisenger/Llama-3.2-3B_2bit_int_7B_teacher

Base model

Finetuned

(54)

this model