Preference-aligned conversational assistant based on Qwen3-1.7B-Base. Fine-tuned using UltraChat and optimized with DPO on UltraFeedback, Intel Orca and Capybara

ollama run ayushshah/qwen3-1.7b-chat:q4_k_m

curl http://localhost:11434/api/chat \
  -d '{
    "model": "ayushshah/qwen3-1.7b-chat:q4_k_m",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='ayushshah/qwen3-1.7b-chat:q4_k_m',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'ayushshah/qwen3-1.7b-chat:q4_k_m',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 3 days ago

3 days ago

17f56e631c78 · 1.1GB ·

model

archqwen3

parameters1.72B

quantizationQ4_K_M

1.1GB

template

{{- range .Messages }}<|im_start|>{{ .Role }} {{ .Content }}<|im_end|> {{ end }}<|im_start|>assistan

102B

system

You are a helpful AI assistant.

31B

params

{ "num_ctx": 32768, "repeat_penalty": 1.05, "stop": [ "<|im_start|>", "<

138B

Readme

Qwen3-1.7B Chat

Qwen3-1.7B Chat is a preference-aligned conversational language model built by fine-tuning Qwen3-1.7B-Base through a two-stage post-training pipeline:

Supervised Fine-Tuning (SFT) on UltraChat
Direct Preference Optimization (DPO) using UltraFeedback, Intel Orca and Capybara

The model is designed to provide more natural, helpful and conversational responses while preserving the strong capabilities of the base model.

Full model, documentation, benchmarks and training details are available on Hugging Face:

Merged Model: https://huggingface.co/ayushshah/Qwen3-1.7B-Chat
LoRA Adapter: https://huggingface.co/ayushshah/Qwen3-1.7B-Chat-LoRA
SFT Model: https://huggingface.co/ayushshah/Qwen3-1.7B-UltraChat-SFT

Notes

This model is intended for research and general conversational AI use. As a 1.7B parameter model, it may still struggle with complex reasoning, factual accuracy and highly specialized domains compared to much larger language models.

# Qwen3-1.7B Chat

Qwen3-1.7B Chat is a preference-aligned conversational language model built by fine-tuning [Qwen3-1.7B-Base](https://huggingface.co/Qwen/Qwen3-1.7B-Base) through a two-stage post-training pipeline:

- **Supervised Fine-Tuning (SFT)** on UltraChat
- **Direct Preference Optimization (DPO)** using UltraFeedback, Intel Orca and Capybara

The model is designed to provide more natural, helpful and conversational responses while preserving the strong capabilities of the base model.

Full model, documentation, benchmarks and training details are available on Hugging Face:

- Merged Model: [https://huggingface.co/ayushshah/Qwen3-1.7B-Chat](https://huggingface.co/ayushshah/Qwen3-1.7B-Chat)
- LoRA Adapter: [https://huggingface.co/ayushshah/Qwen3-1.7B-Chat-LoRA](https://huggingface.co/ayushshah/Qwen3-1.7B-Chat-LoRA)
- SFT Model: [https://huggingface.co/ayushshah/Qwen3-1.7B-UltraChat-SFT](https://huggingface.co/ayushshah/Qwen3-1.7B-UltraChat-SFT)

### Notes
This model is intended for research and general conversational AI use. As a 1.7B parameter model, it may still struggle with complex reasoning, factual accuracy and highly specialized domains compared to much larger language models.

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