jan-hq/trinity-v1

Jan - Discord

Model Description

This model uses the Slerp merge method from the best models on 14th Dec on the OpenLLM Leaderboard:

1. viethq188/LeoScorpius-7B-Chat-DPO
2. GreenNode/GreenNodeLM-7B-v1olet

• base model: GreenNode/GreenNodeLM-7B-v1olet

The yaml config file for this model is here:

slices:
  - sources:
      - model: viethq188/LeoScorpius-7B-Chat-DPO
        layer_range: [0, 32] 
      - model: GreenNode/GreenNodeLM-7B-v1olet
        layer_range: [0, 32]
merge_method: slerp
base_model: GreenNode/GreenNodeLM-7B-v1olet
parameters:
  t:
    - filter: lm_head 
      value: [0.55]
    - filter: embed_tokens
      value: [0.7]
    - filter: self_attn
      value: [0.65, 0.35]
    - filter: mlp
      value:  [0.35, 0.65]
    - filter: layernorm
      value: [0.4, 0.6]
    - filter: modelnorm
      value: [0.6]
    - value: 0.5 # fallback for rest of tensors
dtype: bfloat16

Thank you Undi95 for the secret sauce and Charles Goddard for mergekit.

Prompt template

Work best on:

{system_message}
### Instruction:
{prompt}

### Response:

Run this model

You can run this model using Jan Desktop on Mac, Windows, or Linux.

Jan is an open source, ChatGPT alternative that is:

• 💻 100% offline on your machine: Your conversations remain confidential, and visible only to you.
• 🗂️ An Open File Format: Conversations and model settings stay on your computer and can be exported or deleted at any time.
• 🌐 OpenAI Compatible: Local server on port 1337 with OpenAI compatible endpoints
• 🌍 Open Source & Free: We build in public; check out our Github

About Jan

Jan believes in the need for an open-source AI ecosystem and is building the infra and tooling to allow open-source AIs to compete on a level playing field with proprietary ones.

Jan's long-term vision is to build a cognitive framework for future robots, who are practical, useful assistants for humans and businesses in everyday life.

Jan Model Merger

This is a test project for merging models.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here.

Metric	Value
Avg.	74.8
ARC (25-shot)	72.27
HellaSwag (10-shot)	88.36
MMLU (5-shot)	65.2
TruthfulQA (0-shot)	69.31
Winogrande (5-shot)	82
GSM8K (5-shot)	71.65

Acknowlegement

• mergekit
• DARE
• SLERP
• lm-evaluation-harness