Instructions to use AIMatrix55/qwen-1.5b-stage1-sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AIMatrix55/qwen-1.5b-stage1-sft with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="AIMatrix55/qwen-1.5b-stage1-sft")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("AIMatrix55/qwen-1.5b-stage1-sft")
model = AutoModelForCausalLM.from_pretrained("AIMatrix55/qwen-1.5b-stage1-sft", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use AIMatrix55/qwen-1.5b-stage1-sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AIMatrix55/qwen-1.5b-stage1-sft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AIMatrix55/qwen-1.5b-stage1-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/AIMatrix55/qwen-1.5b-stage1-sft

SGLang

How to use AIMatrix55/qwen-1.5b-stage1-sft with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "AIMatrix55/qwen-1.5b-stage1-sft" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AIMatrix55/qwen-1.5b-stage1-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "AIMatrix55/qwen-1.5b-stage1-sft" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AIMatrix55/qwen-1.5b-stage1-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use AIMatrix55/qwen-1.5b-stage1-sft with Docker Model Runner:
```
docker model run hf.co/AIMatrix55/qwen-1.5b-stage1-sft
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Google Drive link: http://drive.google.com/drive/folders/1xnONQijiHqUKXC-hLmbhzcCurbqgn5Wi

Model Card for qwen-1.5b-stage1-sft

This model is a fine-tuned version of Qwen/Qwen2.5-1.5B-Instruct, developed by the AI Matrix Team (Khushimalik 19, Aman_deva).

It represents Stage 1 (Cold Start) of a two-stage distillation pipeline designed to create a lightweight, edge-deployable "Distilled Reasoner" for automated 5G network fault analysis. This model leverages a DeepSeek-R1 style training approach, learning to output structured reasoning traces (<think> tags) alongside final answers to ensure interpretability and accurate root cause analysis.

Model Details

Developed by: AI Matrix Team
Model type: Causal Language Model (1.5B parameters)
Language(s): English (Telecom Domain)
License: Apache 2.0
Fine-tuning technique: Supervised Fine-Tuning (SFT) via TRL
Intended Use: Edge-cloud compatible inference for 5G log diagnostics and fault reasoning.

Quick Start

This model is optimized for telecom diagnostics. It produces a reasoning trace inside <think> tags before the final answer.

from transformers import pipeline

# Replace with your actual Hugging Face Hub path
model_id = "qwen-1.5b-stage1-sft" 

# Example 5G Fault Query
query = "Analyze the following log: 'RRC Setup Failure: Cause=Congestion'. Determine the root cause and suggest a fix."

generator = pipeline("text-generation", model=model_id, device_map="auto")

# The model expects a user prompt and will generate a <think> block followed by the answer
output = generator([{"role": "user", "content": query}], max_new_tokens=256, return_full_text=False)[0]

print(output["generated_text"])
# Expected output format:
# <think>
# [Step-by-step reasoning about RRC congestion, signaling resources, etc...]
# </think>
# [Final Answer]

Training Procedure

The solution follows a teacher-student distillation paradigm adapted for telecom log reasoning. This specific model covers Stage 1, where the model learns the syntax, structure, and discipline of the reasoning format.

Data Pipeline

Source: netop/TeleLogs (Domain-specific 5G signaling and fault logs).
Preprocessing: Data was cleaned (noise removal) and formatted into ChatML.
Dataset Size: ~2,300 curated samples.
Format: Inputs include <Question>, <Reasoning Trace>, and <Answer>.

Training Hyperparameters

Method: Supervised Fine-Tuning (SFT)
Optimizer: AdamW (8-bit precision)
Epochs: 3
Goal: Cold start—teaching the model "how to think" rather than just what to answer.

Framework Versions

TRL: 0.27.0
Transformers: 4.57.3
PyTorch: 2.6.0
Datasets: 4.5.0
Tokenizers: 0.22.2

Citations

Project Report: AI Matrix Team. (2025). The AI Telco Troubleshooting: Report.

Libraries:

@misc{vonwerra2022trl,
    title        = {{TRL: Transformer Reinforcement Learning}},
    author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
    year         = 2020,
    journal      = {GitHub repository},
    publisher    = {GitHub},
    howpublished = {\url{https://github.com/huggingface/trl}}
}

Downloads last month: -

Safetensors

Model size

2B params

Tensor type

BF16

Model tree for AIMatrix55/qwen-1.5b-stage1-sft

Base model

Qwen/Qwen2.5-1.5B

Finetuned

Qwen/Qwen2.5-1.5B-Instruct