minjaechoi/qwen3p6-35b-a3b-2p07bit-r64

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:2Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 5, 2026Architecture:Transformer Featherless Exclusive Cold

The minjaechoi/qwen3p6-35b-a3b-2p07bit-r64 is a 35.1 billion parameter language model based on the Qwen3.6-35B-A3B architecture, featuring an average of 2.0729 bits per routed expert. This model utilizes BF16 for other weights and stores dequantized weights in BF16 tensors, ensuring compatibility with standard transformers and vLLM libraries. Its primary differentiator is the highly compressed routed expert architecture, making it suitable for efficient deployment while maintaining performance.

Loading preview...

Overview

This model, minjaechoi/qwen3p6-35b-a3b-2p07bit-r64, is an internal research checkpoint derived from the Qwen/Qwen3.6-35B-A3B base model. It features a 35.1 billion parameter architecture with a focus on efficient deployment through advanced quantization techniques. The model incorporates routed experts, achieving an average compression of 2.0729 bits per expert, while all other weights are maintained in BF16 format. This design allows for significant memory and computational savings.

Key Capabilities

  • Efficient Deployment: Achieves substantial model compression with an average of 2.0729 bits per routed expert, making it suitable for environments with limited resources.
  • Standard Compatibility: Weights are stored dequantized in BF16 tensors, ensuring seamless loading and operation with standard libraries like transformers and vLLM.
  • Base Model Performance: Inherits the capabilities and performance characteristics of its robust base model, Qwen3.6-35B-A3B.

Good for

  • Resource-Constrained Applications: Ideal for scenarios where model size and inference speed are critical, such as edge devices or cost-sensitive cloud deployments.
  • Research and Experimentation: Provides a valuable checkpoint for exploring the trade-offs between quantization, model compression, and performance in large language models.
  • Efficient LLM Integration: Developers looking to integrate a powerful 35B parameter model with reduced memory footprint and faster inference times.

Finetunes
1 models