thinkingmachines/Inkling-2-Small
AI model · https://huggingface.co/thinkingmachines/Inkling-2-Small
First seen by Oriel on .
AI summary
Inkling 2 Small is an open-weight multimodal model with 594B total and 17B active parameters, supporting text and images, long context, coding, agents, and general-purpose applications.
- The 48-layer decoder-only transformer uses sparse MoE routing, activating 8 of 384 experts per token, with hybrid local and global attention and an integrated vision transformer.
- The model supports a context window of up to 1M tokens, accepts UTF-8 text and pixel-based images, and generates UTF-8 text across multiple languages and coding languages.
- The recommended NVFP4 checkpoint requires at least 500GB of aggregated VRAM, while MXFP8 requires 800GB; MXFP8 provides only a marginal 1–2 percentage-point evaluation gain.
- Developers can run it with SGLang, vLLM, TokenSpeed, Unsloth, or Hugging Face, and post-train it with Miles, SkyRL, or Snowflake Cortex Training.
- The model can hallucinate, follow instructions imperfectly, degrade in long conversations, and reflect training-data biases, requiring evaluation, safeguards, and human oversight for high-stakes use.
#multimodal models #mixture of experts #open weights #model safety
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: 429, latest observation ; +307 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: 16,918, latest observation ; +16,906 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | thinkingmachines/Inkling-2-Small | https://huggingface.co/thinkingmachines/Inkling-2-Small Read discussion |