w4d4f4k/ qwen3_v1

8 Downloads Updated 3 weeks ago

based on qwen3 3-4B instruct

ollama run w4d4f4k/qwen3_v1

curl http://localhost:11434/api/chat \
  -d '{
    "model": "w4d4f4k/qwen3_v1",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='w4d4f4k/qwen3_v1',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'w4d4f4k/qwen3_v1',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

Name

1 model

Size / Usage

Context

Input

qwen3_v1:latest

2.5GB · 256K context window · Text · 3 weeks ago

qwen3_v1:latest

2.5GB

256K

Text

Readme

AIIA Qwen3 Agent 4B

Qwen3-4B-Instruct-2507 fine-tuned (LoRA, merged) for the AIIA agent
framework — an autonomous PLAN/BUILD-mode code agent with XML-style tool
calling.

What it does

Executes coding and file-system tasks step by step using AIIA’s tool
protocol:

<WriteFile>                                                                                                                                                        
<fileName>fibonacci.py</fileName>
<contentOfFile>def fibonacci(n): ...</contentOfFile>
</WriteFile>

Supported tools include: ReadFile, WriteFile, CreateFile, AppendFile, TreeView, List, Find, Grep, ReplaceLine, Sed, Diff, ExecuteScript, Terminal, createPlan, nextTask, and more. Task lifecycle is signalled with <nextTask>completed</nextTask> / <nextTask>blocked</nextTask>.

Training

Base: Qwen/Qwen3-4B-Instruct-2507 (Apache 2.0)
Method: LoRA r=32 (q/k/v/o/gate/up/down), merged into base weights
Data: ~2,300 real AIIA agent trajectories (file I/O, shell execution, code editing, planning, multi-step task completion)
Quantization: Q4_K_M GGUF (2.5 GB)

Usage

Designed to run inside the AIIA framework, which injects real tool results and drives the task loop. The model stops generation at tool-call boundaries ([Tool Result: stop sequence is built in), so it also behaves sanely in plain chat — it will emit one tool call and wait.

ollama run w4d4f4k/qwen3_v1

Recommended options: temperature 0.3-0.7 for BUILD mode, context 16k.

Limitations

Specialized for AIIA’s XML tool format; not a general chat model (general ability inherited from Qwen3-4B remains, but responses tend toward agent formatting when a system prompt mentions tasks/tools).
Tool results must be provided by the calling framework; the model does not execute anything itself.
English-focused, matching the training trajectories.

Requirements

~3 GB VRAM/RAM for Q4_K_M + context
Runs on stock Ollama, llama.cpp, LM Studio — no custom runtime needed

# AIIA Qwen3 Agent 4B                                                                                                                                              
                                                                                                                                                                   
Qwen3-4B-Instruct-2507 fine-tuned (LoRA, merged) for the **AIIA agent                                                                                              
framework** — an autonomous PLAN/BUILD-mode code agent with XML-style tool                                                                                         
calling.                                                                                                                                                           
                                                                                                                                                                   
## What it does                                                                                                                                                    
                                                                                                                                                                   
Executes coding and file-system tasks step by step using AIIA's tool                                                                                               
protocol:                                                                                                                                                          
                                                                                                                                                                   
```                                                                                                                                                                
<WriteFile>                                                                                                                                                        
<fileName>fibonacci.py</fileName>
<contentOfFile>def fibonacci(n): ...</contentOfFile>
</WriteFile>
```
 
Supported tools include: ReadFile, WriteFile, CreateFile, AppendFile,
TreeView, List, Find, Grep, ReplaceLine, Sed, Diff, ExecuteScript, Terminal,
createPlan, nextTask, and more. Task lifecycle is signalled with
`<nextTask>completed</nextTask>` / `<nextTask>blocked</nextTask>`.
 
## Training
 
- **Base:** Qwen/Qwen3-4B-Instruct-2507 (Apache 2.0)
- **Method:** LoRA r=32 (q/k/v/o/gate/up/down), merged into base weights
- **Data:** ~2,300 real AIIA agent trajectories (file I/O, shell execution,
  code editing, planning, multi-step task completion)
- **Quantization:** Q4_K_M GGUF (2.5 GB)
 
## Usage
 
Designed to run **inside the AIIA framework**, which injects real tool
results and drives the task loop. The model stops generation at tool-call
boundaries (`[Tool Result:` stop sequence is built in), so it also behaves
sanely in plain chat — it will emit one tool call and wait.
 
```bash
ollama run w4d4f4k/qwen3_v1
```
 
Recommended options: `temperature 0.3-0.7` for BUILD mode, context 16k.
 
## Limitations
 
- Specialized for AIIA's XML tool format; not a general chat model
  (general ability inherited from Qwen3-4B remains, but responses tend
  toward agent formatting when a system prompt mentions tasks/tools).
- Tool results must be provided by the calling framework; the model does
  not execute anything itself.
- English-focused, matching the training trajectories.
 
## Requirements
 
- ~3 GB VRAM/RAM for Q4_K_M + context
- Runs on stock Ollama, llama.cpp, LM Studio — no custom runtime needed

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