Unverified80% confidenceFactTime unknown

OmniMerge V4的Q5版本可以支持100K的上下文长度

1

Sources

80%

Confidence

Medium-term (~90 days)

Relevance

6/1/2026

First Seen

Valid until: 8/30/2026

Sources

Qwen3.6 27B三大邪修量化模型实测：代码暴增15.8PP、40B蒸馏、16GB适配

bilibili尚书省说书人

Related Entities

OmniMerge V4model

Related Claims

UnverifiedOmniMerge V4的Q5版本在RTX 5090上运行速度明显快于普通Q4量化版本78% similar UnverifiedOllama默认提供的模型多为4-bit量化版本（Q4_0或Q4_K_M格式）67% similar UnverifiedOmniMerge V4是由三到四个27B模型通过权重混合融合而成的模型66% similar UnverifiedQLoRA结合4-bit量化，使得在消费级GPU上微调70B参数模型成为可能64% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/18012

API

curl https://kongchang.com/api/v1/knowledge/claims/18012

MCP

get_claim(id=18012)