eve-esa
/

EVE-Instruct-Dspark-best

speculative-decoding

Model card Files Files and versions

EVE-Instruct D-Spark Best

Best D-Spark speculative decoding checkpoint for eve-esa/EVE-Instruct, selected after 10 training epochs by validation loss.

Speculative block size: 8
Target layers: 2, 20, 37
Training sequence length: 4096
Validation loss: 0.632488
Validation full accuracy: 0.486818

Downloads last month: 30

Safetensors

Model size

3B params

Tensor type

I64

·

BF16

·

BOOL

·

Inference Providers NEW

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

Model tree for eve-esa/EVE-Instruct-Dspark-best

Base model

eve-esa/EVE-Instruct

Finetuned

(1)

this model