Qwen1.5-22B-Chat-Merge

Qwen1.5-22B-Chat-Merge

--This is a 22b frankenmerge of qwen1.5-14B-Chat created by interleaving layers of qwen1.5-14B-Chat with itself using mergekit.--

Due to the current absence of intermediary-sized models between 14B and 72B in the Qwen1.5 series, I am trying to make some middle-sized models, such as those with 20B+ and 30B+ parameters, through a merging approach. This initiative aims to enable more individual users to maximize the utilization of their hardware capabilities.

-Quantize

GGUF Here：gguf

-Merge Configuration

This yaml below:

dtype: float16
merge_method: passthrough
slices:
- sources:
  - layer_range: [0, 5]
    model: Qwen/Qwen1.5-14B-Chat
- sources:
  - layer_range: [5, 15]
    model: Qwen/Qwen1.5-14B-Chat
- sources:
  - layer_range: [10, 20]
    model: Qwen/Qwen1.5-14B-Chat
- sources:
  - layer_range: [15, 25]
    model: Qwen/Qwen1.5-14B-Chat
- sources:
  - layer_range: [20, 30]
    model: Qwen/Qwen1.5-14B-Chat
- sources:
  - layer_range: [25, 35]
    model: Qwen/Qwen1.5-14B-Chat
- sources:
  - layer_range: [30, 40]
    model: Qwen/Qwen1.5-14B-Chat

-Performance

• Tips:I don't have the capability to conduct benchmark tests, nor can I even use it extensively enough, so my test results might not be accurate.

It has better performance than the 14B version in most of my own tests (subjective) including comprehension, reasoning and coherence and also writing skills. If you believe in this model's performance, feel free to test it out or offer evaluations. Everyone's tests or evaluations are welcome.