Artemisbunbun/ myrdinn-v1.6

1 Download Updated 22 hours ago

ollama run Artemisbunbun/myrdinn-v1.6

curl http://localhost:11434/api/chat \
  -d '{
    "model": "Artemisbunbun/myrdinn-v1.6",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='Artemisbunbun/myrdinn-v1.6',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'Artemisbunbun/myrdinn-v1.6',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

Name

1 model

Size / Usage

Context

Input

myrdinn-v1.6:latest

2.6GB · 32K context window · Text · 22 hours ago

myrdinn-v1.6:latest

2.6GB

32K

Text

Readme

Myrdinn v1.6

Precision reasoning model built on Qwen2.5-3B-Instruct with selective quantization.

What it is

Myrdinn v1.6 is a 3B-parameter model optimized for mathematical reasoning, code generation, and logical
analysis. It uses a mixed-precision quantization strategy that keeps critical layers (output projection and token embeddings) at higher precision while the transformer body runs at Q5_K_M.

Quantization

| Component | Precision |
|—|—|
| Transformer body (36 layers) | Q5_K_M (5.33 bpw) |
| output.weight | Q8_0 (8-bit) |
| token_embd.weight | Q8_0 (8-bit) |
| Norm weights | F32 (lossless) |

Strengths
Mathematical reasoning: step-by-step derivations with proper notation
Code generation: clean, correct implementations with explanations
Logic and analysis: breaks down problems, identifies assumptions
Concise responses: no filler, no roleplay
Usage

   ollama pull Artemisbunbun/myrdinn-v1.6                                                                    
   ollama run Artemisbunbun/myrdinn-v1.6

Base model

Qwen2.5-3B-Instruct by Alibaba Cloud. Quantized from Q8_0 with llama.cpp.

# Myrdinn v1.6                                                                                            
                                                                                                             
   Precision reasoning model built on Qwen2.5-3B-Instruct with selective quantization.                       
                                                                                                             
  - What it is                                                                                             
                                                                                                             
   Myrdinn v1.6 is a 3B-parameter model optimized for mathematical reasoning, code generation, and logical   
 analysis. It uses a mixed-precision quantization strategy that keeps critical layers (output projection and 
 token embeddings) at higher precision while the transformer body runs at Q5_K_M.                            
                                                                                                             
   - Quantization                                                                                           
                                                                                                             
   | Component | Precision |                                                                                 
   |---|---|                                                                                                 
   | Transformer body (36 layers) | Q5_K_M (5.33 bpw) |                                                      
   | output.weight | Q8_0 (8-bit) |                                                                          
   | token_embd.weight | Q8_0 (8-bit) |                                                                      
   | Norm weights | F32 (lossless) |                                                                         
                                                                                                             
   - Strengths                                                                                              
                                                                                                            
   - **Mathematical reasoning**: step-by-step derivations with proper notation                               
   - **Code generation**: clean, correct implementations with explanations                                   
   - **Logic and analysis**: breaks down problems, identifies assumptions                                    
   - **Concise responses**: no filler, no roleplay                                                           
                                                                                                             
   - Usage                                                                                                  
                                                                                                             
   ```bash                                                                                                   
   ollama pull Artemisbunbun/myrdinn-v1.6                                                                    
   ollama run Artemisbunbun/myrdinn-v1.6                                                                     
 ```                                                                                                         
                                                                                                             
 Base model                                                                                                  
                                                                                                             
 Qwen2.5-3B-Instruct by Alibaba Cloud. Quantized from Q8_0 with llama.cpp.

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)