**Editorial Brief**
The post titled “Now that MTP is merged… What’s the best outputs you’re getting on Qwen 3.6 35B on 2x3090s?” sparked discussion among users of the British AI community, particularly those using the newer architecture known as Qwen 3.6 with a 35B parameter size and running it on two NVIDIA 3090 GPUs.
**What Happened:**
Users shared their experiences and current best outputs with this configuration. Some reported improved performance over previous builds, while others noted that recent changes like the merge of MTP (Model Training Pipeline) have affected output quality or speed. The discussion centered around finding a balance between speed and quality in these configurations.
**Why It Matters:**
This post is significant because it reflects ongoing experimentation within the AI community. Users are trying to optimize their setups, which can lead to better performance outcomes for various applications such as text generation and language modeling. Understanding what works best with different parameters and hardware configurations helps researchers and developers fine-tune their models for specific use cases.
**Takeaways:**
– **Performance Variability:** Different users experienced varying levels of improvement or degradation post-MTP merge, highlighting the need for ongoing testing.
– **Parameter Size Impact:** The 35B parameter size still offers substantial computational power but may require adjustments in training and inference pipelines to achieve optimal performance.
– **Hardware Efficiency:** Users are exploring how best to leverage their hardware (2x3090s) with larger models, indicating a focus on optimizing resource utilization for better efficiency.
This exploration of new configurations underscores the dynamic nature of AI development and the importance of community engagement in refining these tools.
