Work

DFlash 2 vs MTP Speculative Decoding on RTX PRO 6000

Year: 2026

Type: article

Abstract: A second model on the card drafting for the first: DFlash 2 vs built-in MTP on a dense 27B NVFP4, one RTX PRO 6000, real prompts, acceptance and joules logged.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.24058744

Author: Schotten, Stefano

Language: English

Topic: Parallel Computing and Optimization Techniques

Subfield: Hardware and Architecture

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7225585066

FWCI:

Cites:

Cited by:

Related to: 10