Qwen1.5-124B-Chat-Merge

--This is a 124b frankenmerge of qwen1.5-72B-Chat created by interleaving layers of qwen1.5-72B-Chat with itself using mergekit.--

Inspired by other frankenmerge models like goliath-120b and miqu-1-120b

-New Version Conming soon

I have recently created another version of 124B frankenmerge qwen1.5 that performs better than this one, especially in terms of logical abilities and comprehension(It has reached a level close to that of proprietary models in some logic puzzles I designed myself.). It has achieved improved performance through the use of a different merge recipe and is about to be uploaded...

-Quantize

GGUF Here:gguf

-Merge Configuration

This yaml below:

dtype: float16
merge_method: passthrough
slices:
- sources:
  - layer_range: [0, 20]
      model: Qwen/Qwen1.5-72B-Chat
- sources:
  - layer_range: [10, 30]
      model: Qwen/Qwen1.5-72B-Chat
- sources:
  - layer_range: [20, 40]
      model: Qwen/Qwen1.5-72B-Chat
- sources:
  - layer_range: [30, 50]
      model: Qwen/Qwen1.5-72B-Chat
- sources:
  - layer_range: [40, 60]
      model: Qwen/Qwen1.5-72B-Chat
- sources:
  - layer_range: [50, 70]
      model: Qwen/Qwen1.5-72B-Chat
- sources:
  - layer_range: [60, 80]
      model: Qwen/Qwen1.5-72B-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 72B version in most of my own tests (subjective) including comprehension, reasoning and coherence. But the improvement doesn't seem as significant as I had imagined (I've only conducted a few tests). If you believe in this model's performance, feel free to test it out or offer evaluations. Everyone's tests or evaluations are welcome.

Downloads last month
20
Safetensors
Model size
125B params
Tensor type
FP16
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.