Instructions to use tiiuae/Falcon-H1-34B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tiiuae/Falcon-H1-34B-Instruct with Transformers:

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

pipe = pipeline("text-generation", model="tiiuae/Falcon-H1-34B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tiiuae/Falcon-H1-34B-Instruct")
model = AutoModelForCausalLM.from_pretrained("tiiuae/Falcon-H1-34B-Instruct")
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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use tiiuae/Falcon-H1-34B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tiiuae/Falcon-H1-34B-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tiiuae/Falcon-H1-34B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tiiuae/Falcon-H1-34B-Instruct

SGLang

How to use tiiuae/Falcon-H1-34B-Instruct 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 "tiiuae/Falcon-H1-34B-Instruct" \
    --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": "tiiuae/Falcon-H1-34B-Instruct",
		"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 "tiiuae/Falcon-H1-34B-Instruct" \
        --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": "tiiuae/Falcon-H1-34B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tiiuae/Falcon-H1-34B-Instruct with Docker Model Runner:
```
docker model run hf.co/tiiuae/Falcon-H1-34B-Instruct
```

Thanks for not grossly overfitting this model.

by phil111 - opened May 25, 2025

Discussion

phil111

May 25, 2025

•

edited May 25, 2025

Other recent models, including Qwen3 34b, focused their training on test boosting tokens like coding, math, and STEM, and are otherwise PROFOUNDLY ignorant when it comes to all other popular domains of knowledge, such as top movies, music, games, and TV shows.

For example, Qwen3 got the entire cast of the most watched Canadian TV show in history (Corner Gas) wrong, and most of the casts of some of the most watched TV shows globally (e.g. Two and a Half Men) wrong. Same goes for the most popular music, movies, games...

While this model did notably better than Qwen3 34b when it comes to broad knowledge. For example, getting about half the cast of Corner Gas correct and all of Two and a Half men correct. And even nearly the entire casts of less popular shows like Home Improvement correct.

This is based on a spot check at https://huggingface.co/spaces/tiiuae/Falcon-H1-playground. But the performance of this model across popular domains of English knowledge is clearly far higher than Qwen3 34b, so thanks again. Keep up the good work.

JingweiZuo

Technology Innovation Institute org May 25, 2025

Thank you very much for your vibe checks! 🚀

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment