Naphula

Slimaki-24B-v1.1-ramplus_tl

Deploy Dedicated

README

License: apache-2.0

Merge Details

Merge Methods

This model was merged using the following merge method:

Reinforced Agent Merging Plus (Tensor-Local)

Note: This merge was heavily inspired by Maginum Cydoms

This model was requested as a way to test Arcee's newest mergekit method on the Slimaki/Cydoms components.

Configuration

yaml

merge_method: ramplus_tl  base_model: B:\24B\models--anthracite-core--Mistral-Small-3.2-24B-Instruct-2506-Text-Onlymodels:    - model: B:\24B\models--anthracite-core--Mistral-Small-3.2-24B-Instruct-2506-Text-Only  - model: B:\24B\models--TheDrummer--Cydonia-24B-v4.3  - model: B:\24B\models--ReadyArt--4.2.0-Broken-Tutu-24b  - model: B:\24B\models--zerofata--MS3.2-PaintedFantasy-v2-24B   - model: B:\24B\models--TheDrummer--Magidonia-24B-v4.3  - model: B:\24B\models--TheDrummer--Precog-24B-v1    - model: B:\24B\models--zerofata--MS3.2-PaintedFantasy-v3-24Bparameters:    epsilon: 0.001  # Increased from 1e-5 to 1e-3 for denser SFT/DPO task vectors    r: 0.25         # Increased from 0.1 to 0.2-0.3 for better SFT behavior preservation    alpha: 0.4      # Increased from 0.2 to 0.4 for enhanced rescalingdtype: float32out_dtype: bfloat16tokenizer:  source: basechat_template: autoname: 🐌 Ślimaki-24B-v1.1-ramplus_tl

Recommended Settings

In particular temp 1 and topnsigma 1.25 seems to help a lot with improving quality but sometimes you need to reswipe.

Some models are more sensitive to Rep Pen settings, 1.12 is another, you may need to adjust higher for smaller models (like 1.4 if overcooked) or lower for smarter ones.

Not sure about adaptive_p that is a newer setting added to kobold

(bolded kobold non-defaults)

Temp 1.0
TopNSigma 1.25
Min-P 0.1
Repetition Penalty 1.08
Top-P 1.0
Top-K 100
Top-A 0
Typical Sampling 1
Tail-Free Sampling 1
Presence Penalty 0
Sampler Seed -1
Rp.Range 360
Rp.Slope 0.7
Smoothing Factor 0
Smoothing Curve 1
DynaTemp 0
Mirostat Mode OFF ("2" enhances creativity but also errors)

New Settings

I have not tested these yet but some users on reddit have reported the following settings work extremely well

min_p: 0.05
adaptive_target: 0.9 (down to 0.65 for more creative)
adaptive_decay: 0.99
repetition_penalty_range: 0
temp and topnsigma: 0.75

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

Naphula

Model Tree

Base

TheDrummer/Cydonia-24B-v4.3

Base

zerofata/MS3.2-PaintedFantasy-v2-24B

Base

TheDrummer/Magidonia-24B-v4.3

Base

ReadyArt/4.2.0-Broken-Tutu-24b

Base

anthracite-core/Mistral-Small-3.2-24B-Instruct-2506-Text-Only

Base

TheDrummer/Precog-24B-v1

Base

zerofata/MS3.2-PaintedFantasy-v3-24B

Merged

this model

Input Modalities

Text

Output Modalities

Text

Supported Functionality

Dedicated EndpointsContainer

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Merge Details

Merge Methods

This model was merged using the following merge method:

Reinforced Agent Merging Plus (Tensor-Local)

Note: This merge was heavily inspired by Maginum Cydoms

This model was requested as a way to test Arcee's newest mergekit method on the Slimaki/Cydoms components.

Configuration

yaml

merge_method: ramplus_tl  base_model: B:\24B\models--anthracite-core--Mistral-Small-3.2-24B-Instruct-2506-Text-Onlymodels:    - model: B:\24B\models--anthracite-core--Mistral-Small-3.2-24B-Instruct-2506-Text-Only  - model: B:\24B\models--TheDrummer--Cydonia-24B-v4.3  - model: B:\24B\models--ReadyArt--4.2.0-Broken-Tutu-24b  - model: B:\24B\models--zerofata--MS3.2-PaintedFantasy-v2-24B   - model: B:\24B\models--TheDrummer--Magidonia-24B-v4.3  - model: B:\24B\models--TheDrummer--Precog-24B-v1    - model: B:\24B\models--zerofata--MS3.2-PaintedFantasy-v3-24Bparameters:    epsilon: 0.001  # Increased from 1e-5 to 1e-3 for denser SFT/DPO task vectors    r: 0.25         # Increased from 0.1 to 0.2-0.3 for better SFT behavior preservation    alpha: 0.4      # Increased from 0.2 to 0.4 for enhanced rescalingdtype: float32out_dtype: bfloat16tokenizer:  source: basechat_template: autoname: 🐌 Ślimaki-24B-v1.1-ramplus_tl

merge_method: ramplus_tl  base_model: B:\24B\models--anthracite-core--Mistral-Small-3.2-24B-Instruct-2506-Text-Onlymodels:    - model: B:\24B\models--anthracite-core--Mistral-Small-3.2-24B-Instruct-2506-Text-Only  - model: B:\24B\models--TheDrummer--Cydonia-24B-v4.3  - model: B:\24B\models--ReadyArt--4.2.0-Broken-Tutu-24b  - model: B:\24B\models--zerofata--MS3.2-PaintedFantasy-v2-24B   - model: B:\24B\models--TheDrummer--Magidonia-24B-v4.3  - model: B:\24B\models--TheDrummer--Precog-24B-v1    - model: B:\24B\models--zerofata--MS3.2-PaintedFantasy-v3-24Bparameters:    epsilon: 0.001  # Increased from 1e-5 to 1e-3 for denser SFT/DPO task vectors    r: 0.25         # Increased from 0.1 to 0.2-0.3 for better SFT behavior preservation    alpha: 0.4      # Increased from 0.2 to 0.4 for enhanced rescalingdtype: float32out_dtype: bfloat16tokenizer:  source: basechat_template: autoname: 🐌 Ślimaki-24B-v1.1-ramplus_tl

Recommended Settings

In particular temp 1 and topnsigma 1.25 seems to help a lot with improving quality but sometimes you need to reswipe.

Some models are more sensitive to Rep Pen settings, 1.12 is another, you may need to adjust higher for smaller models (like 1.4 if overcooked) or lower for smarter ones.

Not sure about adaptive_p that is a newer setting added to kobold

(bolded kobold non-defaults)

Temp 1.0
TopNSigma 1.25
Min-P 0.1
Repetition Penalty 1.08
Top-P 1.0
Top-K 100
Top-A 0
Typical Sampling 1
Tail-Free Sampling 1
Presence Penalty 0
Sampler Seed -1
Rp.Range 360
Rp.Slope 0.7
Smoothing Factor 0
Smoothing Curve 1
DynaTemp 0
Mirostat Mode OFF ("2" enhances creativity but also errors)

New Settings

I have not tested these yet but some users on reddit have reported the following settings work extremely well

min_p: 0.05
adaptive_target: 0.9 (down to 0.65 for more creative)
adaptive_decay: 0.99
repetition_penalty_range: 0
temp and topnsigma: 0.75