minjaechoi/qwen3p6-35b-a3b-nopin-2p00bit-r44

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:2Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 23, 2026Architecture:Transformer Featherless Exclusive Cold

minjaechoi/qwen3p6-35b-a3b-nopin-2p00bit-r44 is an internal research checkpoint based on the Qwen3.6-35B-A3B architecture, featuring 35.1 billion parameters and a 32768-token context length. This model utilizes a unique quantization scheme where routed experts average 2.0000 bits, while other weights remain in BF16 format. It is designed for efficient inference by storing dequantized weights in BF16 tensors, compatible with standard transformers and vLLM libraries.

Loading preview...

Model Overview

minjaechoi/qwen3p6-35b-a3b-nopin-2p00bit-r44 is an internal research checkpoint derived from the Qwen3.6-35B-A3B base model. This 35.1 billion parameter language model is notable for its specialized quantization approach, which aims to optimize efficiency while maintaining compatibility with existing inference frameworks.

Key Characteristics

  • Base Architecture: Built upon the robust Qwen3.6-35B-A3B model, known for its performance in various language understanding and generation tasks.
  • Quantization Scheme: Employs a unique "routed experts" quantization, where these specific components average 2.0000 bits. All other model weights are maintained in BF16 precision.
  • Inference Compatibility: Weights are stored as dequantized BF16 tensors, ensuring seamless loading and operation with standard libraries like transformers and vLLM.
  • Context Length: Supports a substantial context window of 32768 tokens, enabling processing of longer inputs and generating more coherent, extended outputs.

Potential Use Cases

This model is particularly suited for researchers and developers interested in:

  • Efficient LLM Deployment: Exploring the trade-offs and benefits of advanced quantization techniques for large language models.
  • Resource-Constrained Environments: Leveraging the optimized bit-depth for potentially reduced memory footprint and faster inference speeds compared to full-precision models of similar scale.
  • Qwen Ecosystem Integration: Utilizing a quantized variant within existing workflows that are compatible with the Qwen model family.

Finetunes
1 models