Instructions to use megabytes/Falcon-H1-0.5B-Instruct-heretic with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use megabytes/Falcon-H1-0.5B-Instruct-heretic with Transformers:

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

pipe = pipeline("text-generation", model="megabytes/Falcon-H1-0.5B-Instruct-heretic")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("megabytes/Falcon-H1-0.5B-Instruct-heretic")
model = AutoModelForCausalLM.from_pretrained("megabytes/Falcon-H1-0.5B-Instruct-heretic", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use megabytes/Falcon-H1-0.5B-Instruct-heretic with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "megabytes/Falcon-H1-0.5B-Instruct-heretic"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "megabytes/Falcon-H1-0.5B-Instruct-heretic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/megabytes/Falcon-H1-0.5B-Instruct-heretic

SGLang

How to use megabytes/Falcon-H1-0.5B-Instruct-heretic 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 "megabytes/Falcon-H1-0.5B-Instruct-heretic" \
    --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": "megabytes/Falcon-H1-0.5B-Instruct-heretic",
		"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 "megabytes/Falcon-H1-0.5B-Instruct-heretic" \
        --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": "megabytes/Falcon-H1-0.5B-Instruct-heretic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use megabytes/Falcon-H1-0.5B-Instruct-heretic with Docker Model Runner:
```
docker model run hf.co/megabytes/Falcon-H1-0.5B-Instruct-heretic
```

megabytes commited on Jan 20

Commit

bcff599

verified ·

1 Parent(s): 3feff3b

Fixed base model pointer to Instruct

Browse files

Files changed (1) hide show

README.md +172 -172

README.md CHANGED Viewed

@@ -1,173 +1,173 @@
----
-library_name: transformers
-tags:
-- falcon-h1
-- heretic
-- uncensored
-- decensored
-- abliterated
-license: other
-license_name: falcon-llm-license
-license_link: https://falconllm.tii.ae/falcon-terms-and-conditions.html
-base_model: tiiuae/Falcon-H1-0.5B-Base
-inference: true
----
-# This is a decensored version of [tiiuae/Falcon-H1-0.5B-Instruct](https://huggingface.co/tiiuae/Falcon-H1-0.5B-Instruct), made using [Heretic](https://github.com/p-e-w/heretic) v1.1.0
-## Abliteration parameters
-| Parameter | Value |
-| :-------- | :---: |
-| **direction_index** | per layer |
-| **attn.o_proj.max_weight** | 1.49 |
-| **attn.o_proj.max_weight_position** | 23.82 |
-| **attn.o_proj.min_weight** | 1.27 |
-| **attn.o_proj.min_weight_distance** | 20.96 |
-| **mlp.down_proj.max_weight** | 1.47 |
-| **mlp.down_proj.max_weight_position** | 21.57 |
-| **mlp.down_proj.min_weight** | 1.46 |
-| **mlp.down_proj.min_weight_distance** | 12.86 |
-## Performance
-| Metric | This model | Original model ([tiiuae/Falcon-H1-0.5B-Instruct](https://huggingface.co/tiiuae/Falcon-H1-0.5B-Instruct)) |
-| :----- | :--------: | :---------------------------: |
-| **KL divergence** | 0.0954 | 0 *(by definition)* |
-| **Refusals** | 30/100 | 91/100 |
------
-<img src="https://huggingface.co/datasets/tiiuae/documentation-images/resolve/main/falcon_mamba/falcon-h1-logo.png" alt="drawing" width="800"/>
-#  Table of Contents
-0. [TL;DR](#TL;DR)
-1. [Model Details](#model-details)
-2. [Training Details](#training-details)
-3. [Usage](#usage)
-4. [Evaluation](#evaluation)
-5. [Citation](#citation)
-# TL;DR
-# Model Details
-## Model Description
-- **Developed by:** [https://www.tii.ae](https://www.tii.ae)
-- **Model type:** Causal decoder-only
-- **Architecture:** Hybrid Transformers + Mamba architecture
-- **Language(s) (NLP):** English
-- **License:** Falcon-LLM License
-# Training details
-For more details about the training protocol of this model, please refer to the [Falcon-H1 technical blogpost](https://falcon-lm.github.io/blog/falcon-h1/) and [Technical Report](https://arxiv.org/abs/2507.22448).
-# Usage
-Currently to use this model you can either rely on Hugging Face `transformers`, `vLLM` or `llama.cpp` library.
-## Inference
-Make sure to install the latest version of `transformers` or `vllm`, eventually install these packages from source:
-```bash
-pip install git+https://github.com/huggingface/transformers.git
-```
-For vLLM, make sure to install `vllm>=0.9.0`:
-```bash
-pip install "vllm>=0.9.0"
-```
-### 🤗 transformers
-Refer to the snippet below to run H1 models using 🤗 transformers:
-```python
-import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer
-model_id = "tiiuae/Falcon-H1-1B-Base"
-model = AutoModelForCausalLM.from_pretrained(
-  model_id,
-  torch_dtype=torch.bfloat16,
-  device_map="auto"
-)
-# Perform text generation
-```
-### vLLM
-For vLLM, simply start a server by executing the command below:
-```
-# pip install vllm>=0.9.0
-vllm serve tiiuae/Falcon-H1-1B-Instruct --tensor-parallel-size 2 --data-parallel-size 1
-```
-### `llama.cpp`
-You can find all GGUF files compatible with `llama.cpp` under [our official collection](https://huggingface.co/collections/tiiuae/falcon-h1-6819f2795bc406da60fab8df)
-# Evaluation
-Falcon-H1 series perform very well on a variety of tasks, including reasoning tasks.
-| Tasks | Falcon-H1-0.5B | Qwen3-0.6B | Qwen2.5-0.5B | Gemma3-1B | Llama3.2-1B | Falcon3-1B |
-| --- | --- | --- | --- | --- | --- | --- |
-| **General**  | | | | | |
-| BBH | **42.91** | 32.95 | 33.26 | 35.86 | 33.21 | 34.47 |
-| ARC-C | 37.8 | 31.06 | 33.28 | 34.13 | 34.64 | **43.09** |
-| TruthfulQA | 44.12 | **51.65** | 46.19 | 42.17 | 42.08 | 42.31 |
-| HellaSwag | 51.93 | 42.17 | 52.38 | 42.24 | 55.3 | **58.53** |
-| MMLU | **53.4** | 42.98 | 46.07 | 40.87 | 45.93 | 46.1 |
-| **Math**  | | | | | |
-| GSM8k | **68.39** | 42.61 | 38.51 | 42.38 | 44.28 | 44.05 |
-| MATH-500 | **58.4** | 46.0 | 27.8 | 45.4 | 13.2 | 19.8 |
-| AMC-23 | **33.13** | 27.97 | 12.5 | 19.22 | 7.19 | 6.87 |
-| AIME-24 | **3.75** | 2.71 | 0.62 | 0.42 | 1.46 | 0.41 |
-| AIME-25 | **4.38** | 1.67 | 0.21 | 1.25 | 0.0 | 0.21 |
-| **Science**  | | | | | |
-| GPQA | **29.95** | 26.09 | 26.85 | 28.19 | 26.59 | 26.76 |
-| GPQA_Diamond | 27.95 | 25.08 | 24.24 | 21.55 | 25.08 | **31.31** |
-| MMLU-Pro | **31.03** | 16.95 | 18.73 | 14.46 | 16.2 | 18.49 |
-| MMLU-stem | **54.55** | 39.3 | 39.83 | 35.39 | 39.16 | 39.64 |
-| **Code**  | | | | | |
-| HumanEval | **51.83** | 41.46 | 36.59 | 40.85 | 34.15 | 22.56 |
-| HumanEval+ | **45.12** | 37.19 | 32.32 | 37.2 | 29.88 | 20.73 |
-| MBPP | 42.59 | 56.08 | 46.83 | **57.67** | 33.6 | 20.63 |
-| MBPP+ | 33.07 | 47.08 | 39.68 | **50.0** | 29.37 | 17.2 |
-| LiveCodeBench | 7.05 | **9.78** | 2.94 | 5.09 | 2.35 | 0.78 |
-| CRUXEval | **25.75** | 23.63 | 14.88 | 12.7 | 0.06 | 15.58 |
-| **Instruction Following**  | | | | | |
-| IFEval | **72.07** | 62.16 | 32.11 | 61.48 | 55.34 | 54.26 |
-| Alpaca-Eval | 10.79 | 9.59 | 3.26 | **17.87** | 9.38 | 6.98 |
-| MTBench | **7.06** | 5.75 | 4.71 | 7.03 | 6.37 | 6.03 |
-| LiveBench | 20.8 | **27.78** | 14.27 | 18.79 | 14.97 | 14.1 |
-You can check more in detail on our [our release blogpost](https://falcon-lm.github.io/blog/falcon-h1/), detailed benchmarks.
-# Useful links
-- View [our release blogpost](https://falcon-lm.github.io/blog/falcon-h1/).
-- View [our technical report](https://arxiv.org/abs/2507.22448).
-- Feel free to join [our discord server](https://discord.gg/trwMYP9PYm) if you have any questions or to interact with our researchers and developers.
-# Citation
-If the Falcon-H1 family of models were helpful to your work, feel free to give us a cite.
-```
-@article{falconh1,
-    title={Falcon-H1: A Family of Hybrid-Head Language Models Redefining Efficiency and Performance},
-    author={Jingwei Zuo and Maksim Velikanov and Ilyas Chahed and Younes Belkada and Dhia Eddine Rhayem and Guillaume Kunsch and Hakim Hacid and Hamza Yous and Brahim Farhat and Ibrahim Khadraoui and Mugariya Farooq and Giulia Campesan and Ruxandra Cojocaru and Yasser Djilali and Shi Hu and Iheb Chaabane and Puneesh Khanna and Mohamed El Amine Seddik and Ngoc Dung Huynh and Phuc Le Khac and Leen AlQadi and Billel Mokeddem and Mohamed Chami and Abdalgader Abubaker and Mikhail Lubinets and Kacper Piskorski and Slim Frikha},
-    journal = {arXiv preprint arXiv:2507.22448},
-    year={2025}
-}
 ```

+---
+library_name: transformers
+tags:
+- falcon-h1
+- heretic
+- uncensored
+- decensored
+- abliterated
+license: other
+license_name: falcon-llm-license
+license_link: https://falconllm.tii.ae/falcon-terms-and-conditions.html
+base_model: tiiuae/Falcon-H1-0.5B-Instruct
+inference: true
+---
+# This is a decensored version of [tiiuae/Falcon-H1-0.5B-Instruct](https://huggingface.co/tiiuae/Falcon-H1-0.5B-Instruct), made using [Heretic](https://github.com/p-e-w/heretic) v1.1.0
+## Abliteration parameters
+| Parameter | Value |
+| :-------- | :---: |
+| **direction_index** | per layer |
+| **attn.o_proj.max_weight** | 1.49 |
+| **attn.o_proj.max_weight_position** | 23.82 |
+| **attn.o_proj.min_weight** | 1.27 |
+| **attn.o_proj.min_weight_distance** | 20.96 |
+| **mlp.down_proj.max_weight** | 1.47 |
+| **mlp.down_proj.max_weight_position** | 21.57 |
+| **mlp.down_proj.min_weight** | 1.46 |
+| **mlp.down_proj.min_weight_distance** | 12.86 |
+## Performance
+| Metric | This model | Original model ([tiiuae/Falcon-H1-0.5B-Instruct](https://huggingface.co/tiiuae/Falcon-H1-0.5B-Instruct)) |
+| :----- | :--------: | :---------------------------: |
+| **KL divergence** | 0.0954 | 0 *(by definition)* |
+| **Refusals** | 30/100 | 91/100 |
+-----
+<img src="https://huggingface.co/datasets/tiiuae/documentation-images/resolve/main/falcon_mamba/falcon-h1-logo.png" alt="drawing" width="800"/>
+#  Table of Contents
+0. [TL;DR](#TL;DR)
+1. [Model Details](#model-details)
+2. [Training Details](#training-details)
+3. [Usage](#usage)
+4. [Evaluation](#evaluation)
+5. [Citation](#citation)
+# TL;DR
+# Model Details
+## Model Description
+- **Developed by:** [https://www.tii.ae](https://www.tii.ae)
+- **Model type:** Causal decoder-only
+- **Architecture:** Hybrid Transformers + Mamba architecture
+- **Language(s) (NLP):** English
+- **License:** Falcon-LLM License
+# Training details
+For more details about the training protocol of this model, please refer to the [Falcon-H1 technical blogpost](https://falcon-lm.github.io/blog/falcon-h1/) and [Technical Report](https://arxiv.org/abs/2507.22448).
+# Usage
+Currently to use this model you can either rely on Hugging Face `transformers`, `vLLM` or `llama.cpp` library.
+## Inference
+Make sure to install the latest version of `transformers` or `vllm`, eventually install these packages from source:
+```bash
+pip install git+https://github.com/huggingface/transformers.git
+```
+For vLLM, make sure to install `vllm>=0.9.0`:
+```bash
+pip install "vllm>=0.9.0"
+```
+### 🤗 transformers
+Refer to the snippet below to run H1 models using 🤗 transformers:
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+model_id = "tiiuae/Falcon-H1-1B-Base"
+model = AutoModelForCausalLM.from_pretrained(
+  model_id,
+  torch_dtype=torch.bfloat16,
+  device_map="auto"
+)
+# Perform text generation
+```
+### vLLM
+For vLLM, simply start a server by executing the command below:
+```
+# pip install vllm>=0.9.0
+vllm serve tiiuae/Falcon-H1-1B-Instruct --tensor-parallel-size 2 --data-parallel-size 1
+```
+### `llama.cpp`
+You can find all GGUF files compatible with `llama.cpp` under [our official collection](https://huggingface.co/collections/tiiuae/falcon-h1-6819f2795bc406da60fab8df)
+# Evaluation
+Falcon-H1 series perform very well on a variety of tasks, including reasoning tasks.
+| Tasks | Falcon-H1-0.5B | Qwen3-0.6B | Qwen2.5-0.5B | Gemma3-1B | Llama3.2-1B | Falcon3-1B |
+| --- | --- | --- | --- | --- | --- | --- |
+| **General**  | | | | | |
+| BBH | **42.91** | 32.95 | 33.26 | 35.86 | 33.21 | 34.47 |
+| ARC-C | 37.8 | 31.06 | 33.28 | 34.13 | 34.64 | **43.09** |
+| TruthfulQA | 44.12 | **51.65** | 46.19 | 42.17 | 42.08 | 42.31 |
+| HellaSwag | 51.93 | 42.17 | 52.38 | 42.24 | 55.3 | **58.53** |
+| MMLU | **53.4** | 42.98 | 46.07 | 40.87 | 45.93 | 46.1 |
+| **Math**  | | | | | |
+| GSM8k | **68.39** | 42.61 | 38.51 | 42.38 | 44.28 | 44.05 |
+| MATH-500 | **58.4** | 46.0 | 27.8 | 45.4 | 13.2 | 19.8 |
+| AMC-23 | **33.13** | 27.97 | 12.5 | 19.22 | 7.19 | 6.87 |
+| AIME-24 | **3.75** | 2.71 | 0.62 | 0.42 | 1.46 | 0.41 |
+| AIME-25 | **4.38** | 1.67 | 0.21 | 1.25 | 0.0 | 0.21 |
+| **Science**  | | | | | |
+| GPQA | **29.95** | 26.09 | 26.85 | 28.19 | 26.59 | 26.76 |
+| GPQA_Diamond | 27.95 | 25.08 | 24.24 | 21.55 | 25.08 | **31.31** |
+| MMLU-Pro | **31.03** | 16.95 | 18.73 | 14.46 | 16.2 | 18.49 |
+| MMLU-stem | **54.55** | 39.3 | 39.83 | 35.39 | 39.16 | 39.64 |
+| **Code**  | | | | | |
+| HumanEval | **51.83** | 41.46 | 36.59 | 40.85 | 34.15 | 22.56 |
+| HumanEval+ | **45.12** | 37.19 | 32.32 | 37.2 | 29.88 | 20.73 |
+| MBPP | 42.59 | 56.08 | 46.83 | **57.67** | 33.6 | 20.63 |
+| MBPP+ | 33.07 | 47.08 | 39.68 | **50.0** | 29.37 | 17.2 |
+| LiveCodeBench | 7.05 | **9.78** | 2.94 | 5.09 | 2.35 | 0.78 |
+| CRUXEval | **25.75** | 23.63 | 14.88 | 12.7 | 0.06 | 15.58 |
+| **Instruction Following**  | | | | | |
+| IFEval | **72.07** | 62.16 | 32.11 | 61.48 | 55.34 | 54.26 |
+| Alpaca-Eval | 10.79 | 9.59 | 3.26 | **17.87** | 9.38 | 6.98 |
+| MTBench | **7.06** | 5.75 | 4.71 | 7.03 | 6.37 | 6.03 |
+| LiveBench | 20.8 | **27.78** | 14.27 | 18.79 | 14.97 | 14.1 |
+You can check more in detail on our [our release blogpost](https://falcon-lm.github.io/blog/falcon-h1/), detailed benchmarks.
+# Useful links
+- View [our release blogpost](https://falcon-lm.github.io/blog/falcon-h1/).
+- View [our technical report](https://arxiv.org/abs/2507.22448).
+- Feel free to join [our discord server](https://discord.gg/trwMYP9PYm) if you have any questions or to interact with our researchers and developers.
+# Citation
+If the Falcon-H1 family of models were helpful to your work, feel free to give us a cite.
+```
+@article{falconh1,
+    title={Falcon-H1: A Family of Hybrid-Head Language Models Redefining Efficiency and Performance},
+    author={Jingwei Zuo and Maksim Velikanov and Ilyas Chahed and Younes Belkada and Dhia Eddine Rhayem and Guillaume Kunsch and Hakim Hacid and Hamza Yous and Brahim Farhat and Ibrahim Khadraoui and Mugariya Farooq and Giulia Campesan and Ruxandra Cojocaru and Yasser Djilali and Shi Hu and Iheb Chaabane and Puneesh Khanna and Mohamed El Amine Seddik and Ngoc Dung Huynh and Phuc Le Khac and Leen AlQadi and Billel Mokeddem and Mohamed Chami and Abdalgader Abubaker and Mikhail Lubinets and Kacper Piskorski and Slim Frikha},
+    journal = {arXiv preprint arXiv:2507.22448},
+    year={2025}
+}
 ```