Instructions to use biswadeep2004/gpt2-medium-sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use biswadeep2004/gpt2-medium-sft with Transformers:

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

pipe = pipeline("text-generation", model="biswadeep2004/gpt2-medium-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("biswadeep2004/gpt2-medium-sft")
model = AutoModelForCausalLM.from_pretrained("biswadeep2004/gpt2-medium-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 biswadeep2004/gpt2-medium-sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "biswadeep2004/gpt2-medium-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": "biswadeep2004/gpt2-medium-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/biswadeep2004/gpt2-medium-sft

SGLang

How to use biswadeep2004/gpt2-medium-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 "biswadeep2004/gpt2-medium-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": "biswadeep2004/gpt2-medium-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 "biswadeep2004/gpt2-medium-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": "biswadeep2004/gpt2-medium-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use biswadeep2004/gpt2-medium-sft with Docker Model Runner:
```
docker model run hf.co/biswadeep2004/gpt2-medium-sft
```

Qwen3-0.6B-Base SFT

Qwen3-0.6B-Base fine-tuned with supervised fine-tuning (SFT) on banghua/DL-SFT-Dataset. Built as a learning exercise to compare the raw base model's output before and after SFT.

Model Details

Developed by: biswadeep2004
Model type: Causal language model (Qwen3, 0.6B params)
Language(s): Mostly English
License: Apache 2.0 (inherited from the base model)
Finetuned from: Qwen/Qwen3-0.6B-Base

Intended Use

Experimentation and learning about SFT. Not intended for production use or for factual or safety-critical answers.

Limitations

Small model trained on a small dataset (~3k examples). SFT teaches response format and style, not new knowledge, so it can hallucinate and make arithmetic or factual mistakes.

How to Use

from transformers import AutoTokenizer, AutoModelForCausalLM

repo = "biswadeep2004/<your-repo-name>"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo)

msgs = [{"role": "user", "content": "Why is the sky blue?"}]
prompt = tok.apply_chat_template(msgs, tokenize=False, add_generation_prompt=True)
inputs = tok(prompt, return_tensors="pt", add_special_tokens=False)
out = model.generate(**inputs, max_new_tokens=200, do_sample=False)
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Training Details

Data: banghua/DL-SFT-Dataset (single-turn user/assistant conversations)
Format: ChatML (<|im_start|> / <|im_end|>), with loss computed on assistant tokens only (assistant_only_loss=True)
Method: Full fine-tuning with TRL SFTTrainer
Hyperparameters: learning rate 1e-5 (cosine, 20 warmup steps), 2 epochs, per-device batch size 1 × 16 gradient accumulation, max length 1024, fp16 mixed precision, gradient checkpointing
Hardware: 1× NVIDIA T4 (Google Colab)
Final training loss:

Evaluation

Qualitative only: informal before/after comparison on a handful of prompts. No benchmark scores.

Downloads last month: 208

Safetensors

Model size

0.6B params

Tensor type

F32

Model tree for biswadeep2004/gpt2-medium-sft

Base model

Qwen/Qwen3-0.6B-Base

Finetuned

(771)

this model

biswadeep2004
/

gpt2-medium-sft