Instructions to use skyeflo/qwen3-decodable-story-sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use skyeflo/qwen3-decodable-story-sft with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("unsloth/Qwen3-0.6B-unsloth-bnb-4bit")
model = PeftModel.from_pretrained(base_model, "skyeflo/qwen3-decodable-story-sft")

Transformers

How to use skyeflo/qwen3-decodable-story-sft with Transformers:

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

pipe = pipeline("text-generation", model="skyeflo/qwen3-decodable-story-sft")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("skyeflo/qwen3-decodable-story-sft", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use skyeflo/qwen3-decodable-story-sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "skyeflo/qwen3-decodable-story-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": "skyeflo/qwen3-decodable-story-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/skyeflo/qwen3-decodable-story-sft

SGLang

How to use skyeflo/qwen3-decodable-story-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 "skyeflo/qwen3-decodable-story-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": "skyeflo/qwen3-decodable-story-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 "skyeflo/qwen3-decodable-story-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": "skyeflo/qwen3-decodable-story-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Desktop
Docker Model Runner
How to use skyeflo/qwen3-decodable-story-sft with Docker Model Runner:
```
docker model run hf.co/skyeflo/qwen3-decodable-story-sft
```

qwen3-decodable-story-sft / README.md

skyeflo

Update README.md

45dc42b verified 3 months ago

preview code

Raw History Blame Contribute Delete

2.47 kB

	---
	base_model: unsloth/Qwen3-0.6B-unsloth-bnb-4bit
	library_name: peft
	pipeline_tag: text-generation
	tags:
	- base_model:adapter:unsloth/Qwen3-0.6B-unsloth-bnb-4bit
	- lora
	- sft
	- transformers
	- trl
	- unsloth
	---

	# Qwen3 Decodable Story SFT

	## Overview

	Qwen3 Decodable Story SFT is a QLoRA fine-tuned version of Qwen3-0.6B developed to generate decodable reading stories for beginning readers.

	Given one or more target phonics patterns, the model generates a short story intended to emphasize those patterns while maintaining a coherent beginning, middle, and ending.

	This project explores whether supervised fine-tuning on a relatively small, curated dataset can teach a reliable educational behavior to a small language model.

	## Model Details

	- Base model: `unsloth/Qwen3-0.6B-unsloth-bnb-4bit`
	- Fine-tuning method: QLoRA (Unsloth)
	- Training framework: TRL SFTTrainer
	- Task: Decodable story generation

	## Supported Phonics Patterns

	- Short-vowel words
	- Consonant blends
	- Consonant digraphs
	- Final-e words
	- Vowel teams
	- R-controlled vowels
	- Diphthongs
	- Multisyllabic words

	## Evaluation

	The model was evaluated against the base model using held-out prompts.

	Evaluation consisted of:

	- Objective phonics-pattern compliance metrics
	- Blind LLM-as-a-judge scoring
	- Qualitative error analysis

	The fine-tuned model demonstrated improved adherence to the requested phonics behavior compared with the base model while continuing to exhibit limitations on more complex phonics combinations.

	## Intended Use

	This model is intended for:

	- Educational NLP experiments
	- Decodable story generation
	- Small language model fine-tuning demonstrations
	- Research on behavior-specific supervised fine-tuning

	## Limitations

	This is a research prototype and is not intended for instructional or production use.

	Performance is stronger on simpler phonics patterns than on prompts requiring multiple or more complex phonics constraints. Additional curated training data would likely improve consistency.

	## Example Prompt

	```
	Write a decodable story.

	Emphasize these phonics patterns:
	- final-e words
	- consonant blends

	Example words containing these patterns:
	- final-e words: cake, game, bike, time, home
	- consonant blends: stop, frog, black, plant, grin

	Write a coherent narrative with a clear beginning, middle, and ending.

	Include a natural story title.

	Do not mention phonics terms or pattern names in the title or story.
	```