Jellon/Moonlight-L3-15B-v2.5-64k-exl2-4bpw

4bpw exl2 quant of: https://huggingface.co/Darkknight535/Moonlight-L3-15B-v2.5-64k

The model is a bit buggy and tends to runaway, I recommend you use:

• Temperature: 0.3-0.6
• Min_P: 0.15
• Repetition Penalty: 1.05

---

MoonLight-L3-15B-V2.5-64K

---

• Enhancements:
  • Merging: Merge with lunaris to fix some issues. Note : This model is experimental so feedback is needed.

---

Quants (Thanks to mradermacher)

Imatrix Normal

Instruct Template

Default llama3 instruct and context preset, but here is the one i use. Instruct Context

Samplers

Creative

Temp : 1.2
Min P : 0.01

Repetition Penalty : 1.05
OR
Dry 0.7, 1.75, 2,0
XTC 0.1, 0.5

[And everything else neutral]

Normal(Precise)

Temp : 0.6 - 0.8
Min P : 0.1

Repetition Penalty : 1.1
OR
Dry 1, 1.75, 2,0
XTC 0.1, 0.2

[And everything else neutral]

System Prompt

Develop the plot slowly, always stay in character. Describe all actions in full, elaborate, explicit, graphic, and vivid detail. Mention all relevant sensory perceptions. Keep the story immersive and engaging. Speak as other person when needed and prefix with the name of person you're speaking as except {{user}}.

FeedBack

FeedBack here

merge

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the SLERP merge method.

Models Merged

The following models were included in the merge:

• Darkknight535/Moonlight-L3-15B-v2-64k
• Tremontaine/L3-Lunaris-v1-15B

Configuration

The following YAML configuration was used to produce this model:

slices:
  - sources:
      - model: Darkknight535/Moonlight-L3-15B-v2-64k
        layer_range: [0, 64]
      - model: Tremontaine/L3-Lunaris-v1-15B
        layer_range: [0, 64]
        
merge_method: slerp
base_model: Darkknight535/Moonlight-L3-15B-v2-64k
parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5 # fallback for rest of tensors
dtype: bfloat16