Instructions to use open-machine/Qwen3-8B-FlashNorm with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use open-machine/Qwen3-8B-FlashNorm with Transformers:

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

pipe = pipeline("text-generation", model="open-machine/Qwen3-8B-FlashNorm")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("open-machine/Qwen3-8B-FlashNorm")
model = AutoModelForCausalLM.from_pretrained("open-machine/Qwen3-8B-FlashNorm")
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 open-machine/Qwen3-8B-FlashNorm with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "open-machine/Qwen3-8B-FlashNorm"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "open-machine/Qwen3-8B-FlashNorm",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/open-machine/Qwen3-8B-FlashNorm

SGLang

How to use open-machine/Qwen3-8B-FlashNorm 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 "open-machine/Qwen3-8B-FlashNorm" \
    --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": "open-machine/Qwen3-8B-FlashNorm",
		"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 "open-machine/Qwen3-8B-FlashNorm" \
        --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": "open-machine/Qwen3-8B-FlashNorm",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use open-machine/Qwen3-8B-FlashNorm with Docker Model Runner:
```
docker model run hf.co/open-machine/Qwen3-8B-FlashNorm
```

Add library_name and paper link

by nielsr HF Staff - opened Apr 23

base: refs/heads/main

←

from: refs/pr/1

Discussion Files changed

+26

-9

Files changed (1) hide show

README.md +26 -9

README.md CHANGED Viewed

@@ -1,17 +1,20 @@
 ---
-license: apache-2.0
 base_model: Qwen/Qwen3-8B
-tags:
-  - flashnorm
-  - transformer-tricks
-  - efficient-inference
-  - weightless-rmsnorm
 pipeline_tag: text-generation
 ---
 # Qwen3-8B-FlashNorm
-FlashNorm-prepared checkpoint of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B). Mathematically equivalent to the source model. The per-channel RMSNorm weight tensors (`input_layernorm.weight`, `post_attention_layernorm.weight`, `model.norm.weight`) are folded into the following linear layers and then removed from the state dict entirely.
 > **Framework support note.** Stock vLLM currently does not load this checkpoint because the norm weight tensors are absent. The upstream patch to accept missing tensors is tracked at: **TBD (vLLM issue link)**. Until the patch lands, use HuggingFace Transformers; it loads this with a warning that norm weights were not initialized and defaults them to ones, which is the correct behavior for FlashNorm.
@@ -23,7 +26,7 @@ An exact reformulation of `RMSNorm -> Linear`:
 - After folding, the RMSNorm layer has no learnable per-channel scale. At runtime it simply divides by `rms(x)`.
 - The resulting model computes the same output as the original, by Proposition 1 of the FlashNorm paper.
-See the [paper](https://arxiv.org/abs/2407.09577) and the [transformer-tricks](https://github.com/OpenMachine-ai/transformer-tricks) repo for details.
 ## Usage
@@ -53,6 +56,20 @@ A warning about missing norm weights is expected; Transformers defaults those to
 Not yet supported. See the tracking issue linked above.
 ## License
-Inherited from the source model.

 ---
 base_model: Qwen/Qwen3-8B
+license: apache-2.0
 pipeline_tag: text-generation
+library_name: transformers
+tags:
+- flashnorm
+- transformer-tricks
+- efficient-inference
+- weightless-rmsnorm
 ---
 # Qwen3-8B-FlashNorm
+This is a FlashNorm-prepared checkpoint of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B), as presented in the paper [FlashNorm: Fast Normalization for Transformers](https://huggingface.co/papers/2407.09577).
+Mathematically equivalent to the source model. The per-channel RMSNorm weight tensors (`input_layernorm.weight`, `post_attention_layernorm.weight`, `model.norm.weight`) are folded into the following linear layers and then removed from the state dict entirely.
 > **Framework support note.** Stock vLLM currently does not load this checkpoint because the norm weight tensors are absent. The upstream patch to accept missing tensors is tracked at: **TBD (vLLM issue link)**. Until the patch lands, use HuggingFace Transformers; it loads this with a warning that norm weights were not initialized and defaults them to ones, which is the correct behavior for FlashNorm.
 - After folding, the RMSNorm layer has no learnable per-channel scale. At runtime it simply divides by `rms(x)`.
 - The resulting model computes the same output as the original, by Proposition 1 of the FlashNorm paper.
+See the [paper](https://huggingface.co/papers/2407.09577) and the [transformer-tricks](https://github.com/OpenMachine-ai/transformer-tricks) repo for details.
 ## Usage
 Not yet supported. See the tracking issue linked above.
+## Citation
+```bibtex
+@misc{graef2024flashnormfastnormalizationtransformers,
+      title={FlashNorm: Fast Normalization for Transformers},
+      author={Nils Graef and Matthew Clapp and Andrew Wasielewski},
+      year={2024},
+      eprint={2407.09577},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2407.09577},
+}
+```
 ## License
+Inherited from the source model.