minjaechoi/qwen36-35b-a3b-2p00bit-r5

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:2Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 17, 2026Architecture:Transformer Featherless Exclusive Cold

The minjaechoi/qwen36-35b-a3b-2p00bit-r5 model is an internal research checkpoint based on the Qwen/Qwen3.6-35B-A3B architecture, featuring 35.1 billion parameters and a 32768-token context length. This model utilizes routed experts with an average quantization of 2.00 bits, while other weights are stored in BF16. It is designed for efficient loading with standard `transformers` and vLLM libraries, focusing on optimized performance through mixed-precision quantization.

Loading preview...

Model Overview

The minjaechoi/qwen36-35b-a3b-2p00bit-r5 is an internal research checkpoint derived from the Qwen/Qwen3.6-35B-A3B base model. This large language model incorporates a unique quantization strategy to optimize its performance and efficiency.

Key Characteristics

  • Base Architecture: Built upon the robust Qwen3.6-35B-A3B model, providing a strong foundation for its capabilities.
  • Parameter Count: Features 35.1 billion parameters, indicating a substantial capacity for complex language tasks.
  • Context Length: Supports a generous context window of 32768 tokens, enabling the processing of longer inputs and maintaining coherence over extended conversations or documents.
  • Quantization Strategy: Employs "routed experts" with an average quantization of 2.00 bits. This technique aims to reduce the model's memory footprint and computational requirements while preserving performance.
  • Mixed Precision: While routed experts are quantized, other weights are maintained in BF16 (Brain Floating Point) format, balancing efficiency with precision.
  • Compatibility: Designed for seamless integration and loading with standard machine learning libraries such as transformers and vLLM, ensuring ease of use for developers.

Intended Use Cases

This model is particularly suitable for research and development efforts focused on:

  • Efficient LLM Deployment: Exploring methods to deploy large language models with reduced resource consumption.
  • Performance Optimization: Investigating the impact of advanced quantization techniques on model inference speed and memory usage.
  • Experimentation with Routed Experts: Utilizing a model that specifically implements and tests the routed expert paradigm for conditional computation.

Its license follows that of the base Qwen/Qwen3.6-35B-A3B model.

Finetunes
1 models