---
title: "Unsloth's Qwen3.5-122B Runs 2x Faster on Local Hardware Without a Draft Model"
subtitle: "Unsloth ships GGUF quants of Qwen3.5-122B-A10B with built-in Multi-Token Prediction, delivering 1.5-2x faster local inference via llama.cpp."
url: "https://alphasignal.ai/news/unsloth-s-qwen3-5-122b-runs-2x-faster-on-local-hardware-without-a-draft-model"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "Unsloth AI"
company_url: "https://alphasignal.ai/company/unslothai"
published: "2026-05-15T19:03:18.000Z"
updated: "2026-08-22T02:00:33.097Z"
type: "Model"
topics:
  - "Llms"
  - "Open Source"
subtopics:
  - "Mixture Of Experts"
  - "Vision Language"
  - "Long Context"
upvotes: 2855
source_url: "https://huggingface.co/unsloth/Qwen3.5-122B-A10B-MTP-GGUF"
comments: 0
truncated: true
---

# Unsloth's Qwen3.5-122B Runs 2x Faster on Local Hardware Without a Draft Model

> Unsloth ships GGUF quants of Qwen3.5-122B-A10B with built-in Multi-Token Prediction, delivering 1.5-2x faster local inference via llama.cpp.

Running a 122B-parameter multimodal model on your own hardware just got noticeably faster. [Unsloth's new MTP-enabled GGUF release](https://huggingface.co/unsloth/Qwen3.5-122B-A10B-MTP-GGUF) of Qwen3.5-122B-A10B pairs Alibaba's mid-tier flagship with Multi-Token Prediction speculative decoding that llama.cpp only started supporting a few days earlier, and the combination is producing 1.5-2x throughput gains for local users.

The headline change is baked into the filename. Standard GGUF quants of Qwen3.5 already exist, but this variant preserves the model's MTP layers so llama.cpp can use them as a self-drafting speculator. [The upstream PR](https://github.com/ggml-org/llama.cpp/pull/22673) that made this possible was merged into llama.cpp on May 16, and [Unsloth notes](https://huggingface.co/unsloth/Qwen3.5-122B-A10B-MTP-GGUF) that MTP speculative decoding delivers roughly 1.5-2x faster generation once you build llama.cpp from the MTP PR branch.

## Self-drafting, no second model required

Speculative decoding normally requires a smaller draft model running alongside the main one. MTP folds that draft into the base model itself: extra heads predict several tokens ahead in a single forward pass, and the main model verifies them in parallel. If the guesses match, you emit multiple tokens per step instead of one. Reports from NVFP4 deployments on DGX Spark hardware suggest MTP acceptance rates can approach 100% on suitable workloads, which is where the throughput multiplier comes from.

_Preview only. Full text requires an AlphaSignal Pro account._

## Sources

- [unsloth/Qwen3.5-122B-A10B-MTP-GGUF](https://huggingface.co/unsloth/Qwen3.5-122B-A10B-MTP-GGUF)
- [Qwen/Qwen3.5-122B-A10B](https://huggingface.co/Qwen/Qwen3.5-122B-A10B)
- [llama.cpp MTP PR #22673](https://github.com/ggml-org/llama.cpp/pull/22673)
- [Qwen3.5-122B-A10B Specifications](https://apxml.com/models/qwen35-122b-a10b)
- [SPARK_Qwen3.5-122B-A10B-NVFP4 Release Notes](https://github.com/bjk110/SPARK_Qwen3.5-122B-A10B-NVFP4/releases)