Qwen1.5 0.5B Chat
The 0.5B-parameter chat model of Qwen1.5, the smallest size in the series, post-trained with supervised fine-tuning and DPO. It suits short replies and rewrites where the prompt supplies the facts.
Overview
Qwen1.5-0.5B-Chat is the aligned chat model at the smallest size of Qwen1.5, a release Qwen describes as the beta version of Qwen2. The series is a decoder-only Transformer line in eight sizes: dense models at 0.5B, 1.8B, 4B, 7B, 14B, 32B and 72B, plus a 14B mixture-of-experts model with 2.7B parameters active. Qwen published a base model and a chat model at each size, and this page serves the chat one.
Qwen pretrained the models on a large amount of data, then post-trained the chat versions with supervised fine-tuning and direct preference optimization. The card lists SwiGLU activation and attention QKV bias in the architecture, a tokenizer adapted to several natural languages and to code, and multilingual support in both the base and chat models.
What you can build
- A research prototype that answers simple questions from text you paste into the prompt, such as a short FAQ, to see how far a 0.5B model gets.
- Rewriting experiments that shorten a paragraph or shift its tone.
- An evaluation run that sorts short sample messages into labels you define in the system prompt, to measure classification quality at this size.
- Prototypes of multi-turn flows, to test history handling and system prompts before the flow moves to a larger model.
Inputs
The prompt is the only required field. The system prompt comes prefilled with a generic helpful-assistant instruction; replace it with your own role and format rules.
Two optional IDs control memory. The user_id field ties stored chat history to one of your users, and session_id splits that user's history into separate conversations. Keep both the same to continue a thread, or change session_id to start a fresh one.
Sampling starts at temperature 0.7 (range 0 to 5), top_p 0.95, top_k 0 and repetition_penalty 1.0, and length_penalty, max_tokens, min_tokens and semicolon-separated stop_sequences are there too. The seed defaults to 123456. The quantization switch is on by default: it runs the weights at reduced precision, which the page's own note says can slightly affect output quality.
Outputs
One plain-text reply per request. The sample on this page answers a question about the Eiffel Tower in a single paragraph of English.
Recommended settings
Qwen's card has one tip for this model: when replies switch language mid-sentence or go wrong in other ways, use the values from its generation_config.json. That file sets top_p to 0.8 and repetition_penalty to 1.1, with sampling on. This page starts at 0.95 and 1.0, so change both by hand.
Limitations
The sample output shows how a 0.5B model handles recall. It dates the Eiffel Tower to 1886, says the tower is made of stone and calls it the tallest man-made structure in the world, and a Chinese word turns up in the middle of the English text. Put the facts the model needs into the prompt and check anything it states from memory.
This page takes text and returns text. There are no fields for tool calls or JSON output.
Safety & compliance
Qwen1.5-0.5B-Chat is released under the Tongyi Qianwen Research License Agreement. It allows the model to be used and modified for non-commercial purposes only, which the license defines as research or evaluation, and commercial use requires a separate license from Alibaba Cloud. The license also forbids using the model or its outputs to improve any other large language model, with Tongyi Qianwen and its derivatives excepted.
API quick start
Run Qwen1.5-0.5B-Chat with a single API call.
{
"prompt": "What are some interesting historical even…",
"user_id": "...",
"session_id": "...",
"system_prompt": "You are a helpful, respectful and honest …"
}curl -X POST "https://api.wiro.ai/v1/Run/qwen/qwen1-5-0-5b-chat" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "What are some interesting historical even…",
"user_id": "...",
"session_id": "...",
"system_prompt": "You are a helpful, respectful and honest …"
}
JSONOther models you might like
Explore similar llm & chat models.



