Oriel — AI & technology worth noticing, every day.

XHToken/Spark-X2.6-4B

AI model · https://huggingface.co/XHToken/Spark-X2.6-4B

First seen by Oriel on .

AI summary

Spark-X2.6-4B is a post-trained 4B-parameter model for on-device agents, claiming stronger coding, tool use, long-context, multilingual, and reasoning performance without changing Spark-X2.5-4B’s architecture.

  • The model offers configurable thinking levels of none, low, and high to trade reasoning quality against latency and inference cost.
  • The authors report best-in-class results among similarly sized open-source models and performance matching or exceeding roughly 9B models on several benchmarks.
  • The repository provides deployment instructions for Transformers, SGLang, vLLM, MLX, llama.cpp, Ollama, LM Studio, and Unsloth across supported GPU, NPU, CPU, and Apple silicon setups.
  • SGLang examples configure a maximum context length of 1,048,576 tokens, but the documentation says this requires sufficient device memory.
  • The model series is licensed under Apache 2.0, and Llama-Factory is recommended for fine-tuning.

#language models #on-device ai #llm agents #model inference

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Models: 1,003, latest observation ; +884 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Models: 19,588, latest observation ; +19,503 over observed period since

Aliases

SourceExternal IDURL
Hugging Face ModelsXHToken/Spark-X2.6-4Bhttps://huggingface.co/XHToken/Spark-X2.6-4B Read discussion