Unverified50% confidenceFactExact time
该模型在 huihui-ai/Huihui-Qwen3.8-27B-abliterated 基础模型之上训练了一个 rank-256 的 LoRA 适配器
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverified华为昇腾系列AI芯片采用自研达芬奇（DaVinci）架构，昇腾910系列定位高性能训练芯片，昇腾310系列主攻边缘推理场景66% similarUnverified该基准测试在Gemma、Qwen、Kimi、MiniMax四类模型上，基于AMD Instinct MI300X与MI355X硬件进行62% similarUnverifiedAdaptiveSpec支持DeepSeek-R1-Distill-Llama-8B、Llama-3.1-8B-Instruct和Qwen3-8B三个主流目标模型62% similarUnverified需要4K H.265硬件转码时，选择带Intel核显（如J4125/N100）或Celeron的机型，ARM架构（如瑞昱RTD1296）仅支持部分格式硬解，遇到不兼容编码会CPU软解卡顿61% similarUnverifiedOrnith 35B是Deep Reinforce AI在Gemma和Qwen 3.5基础上进行后训练推出的模型61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/913191API
curl https://kongchang.com/api/v1/knowledge/claims/913191MCP
get_claim(id=913191)