Hugging Face 发布 Inkling 2 Mini 循环层 MoE 模型
Hugging Face 宣布 Thinking Machines Lab 的 Inkling 2 Mini 上线,称其为多模态 MoE 模型,总参数约 260 亿,每 token 激活参数约 30 亿。其 24 层解码器中的 12 个中间层会循环运行三次,等效于 48 次逻辑层访问,并支持 100 万 token 上下文与图像、文本输入。模型以 NVFP4 和 MXFP8 专家权重的混合精度形式发布,并附带 DSpark drafter,transformers、Llama App、llama.cpp、MLX 与 SGLang 均已支持。Hugging Face 称在非正式测试中其表现至少与 Qwen 3.8 27B 相当,但同时表示仍需更多验证。
这种循环层的多模态 MoE 为本地与数据中心用户提供了一个支持长上下文和图像输入的新选择,但其与 Qwen 3.8 27B 的对比仍属非正式、未经验证的测试。
Today we are releasing Inkling 2 Mini, a fast, powerful model that’s small enough to run on your laptop.
Inkling 2 Mini is a looped transformer, running the middle layers of the model three times for every token. This allows the model to apply more reasoning and reach the right answer with fewer tokens, especially on tasks such as agentic coding that depend on reasoning quality. Inkling 2 Mini leads every open-weights model with less than 100B total parameters on benchmarks such as DeepSWE v1.1 (48.2%) and Vals Index v2.2 (46.5%). It runs faster than similarly performing models as well.
DeepSWE v1.1
SWE-bench Pro v2 Hard (51 tasks)
Reported coding scores for models under 100B total parameters. The green line is the frontier with Inkling 2 Mini, the dashed line the frontier without it, and the shading the ground it adds. Decoding speed is a theoretical ceiling at 614 GB/s with BF16 weights and batch size one, including repeated reads across loops. Evaluation setups vary by model.
Inkling 2 Mini is a Mixture-of-Experts model with 26.4B total parameters, 3.4B active, and 5.4B weight activations per token to account for looping. Looping allows the model to reason longer without writing its reasoning down, which matters for monitoring: our ability to oversee and understand what the model does by reading its chain of thought (CoT).
部分来源内容未展示,可前往原文继续阅读。
#Inkling 2 Mini#Thinking Machines Lab#Hugging Face#Qwen 3.8 27B#SGLang#Llama App#llama.cpp#MLX#transformers#DSpark#Nanbeige