README.md

---
base_model: Qwen/Qwen3-8B
tags:
- transformers
- torchao
- qwen3
license: apache-2.0
language:
- en
---

This repository hosts the **Qwen3-8B** model quantized with [torchao](https://huggingface.co/docs/transformers/main/en/quantization/torchao) 
using int4 weight-only quantization and the [awq](https://arxiv.org/abs/2306.00978) algorithm.
This work is brought to you by the PyTorch team. This model can be used directly or served using [vLLM](https://docs.vllm.ai/en/latest/) for 53% VRAM reduction (7.82 GB needed) 
and 1.34x speedup on H100 GPUs for batch size 1. The model is calibrated with 10 samples from `mmlu_abstract_algebra` task to recover the accuracy for `mmlu_abstract_algebra` specifically.
AWQ-INT4 improves the accuracy of `mmlu_abstract_algebra` of INT4 from 55 to 56, while the bfloat16 baseline is 58.

# Inference with vLLM
Install vllm nightly and torchao nightly to get some recent changes:
```
pip install vllm --pre --extra-index-url https://wheels.vllm.ai/nightly
pip install torchao
```

## Serving
Then we can serve with the following command:
```Shell
# Server
export MODEL=pytorch/Qwen3-8B-AWQ-INT4
VLLM_DISABLE_COMPILE_CACHE=1 vllm serve $MODEL --tokenizer $MODEL -O3
```

```Shell
# Client
curl http://localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{
  "model": "pytorch/Qwen3-8B-AWQ-INT4",
  "messages": [
    {"role": "user", "content": "Give me a short introduction to large language models."}
  ],
  "temperature": 0.6,
  "top_p": 0.95,
  "top_k": 20,
  "max_tokens": 32768
}'
```

Note: please use `VLLM_DISABLE_COMPILE_CACHE=1` to disable compile cache when running this code, e.g. `VLLM_DISABLE_COMPILE_CACHE=1 python example.py`, since there are some issues with the composability of compile in vLLM and torchao,
this is expected be resolved in pytorch 2.8.

# Inference with Transformers

Install the required packages:
```Shell
pip install git+https://github.com/huggingface/transformers@main
pip install torchao
pip install torch
pip install accelerate
```

Example:
```Py
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "pytorch/Qwen3-8B-AWQ-INT4"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

# prepare the model input
prompt = "Give me a short introduction to large language model."
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True # Switches between thinking and non-thinking modes. Default is True.
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()

# parsing thinking content
try:
    # rindex finding 151668 (</think>)
    index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
    index = 0

thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("
")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("
")

print("thinking content:", thinking_content)
print("content:", content)
```




# Quantization Recipe

Install the required packages:
```Shell
pip install git+https://github.com/huggingface/transformers@main
pip install --pre torchao --index-url https://download.pytorch.org/whl/nightly/cu126
pip install torch
pip install accelerate
```



Use the following code to get the quantized model:
```Py
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TorchAoConfig

model_id = "Qwen/Qwen3-8B"
model_to_quantize = "Qwen/Qwen3-8B"


from torchao.quantization import Int4WeightOnlyConfig, quantize_
from torchao.prototype.awq import (
    AWQConfig,
)
from torchao._models._eval import TransformerEvalWrapper
model = AutoModelForCausalLM.from_pretrained(
    model_to_quantize,
    device_map="auto",
    torch_dtype=torch.bfloat16,
)
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Note: this is only compatible with H100
base_config = Int4WeightOnlyConfig(group_size=128)
# for A100, please use the following for base_config:
# base_config = Int4WeightOnlyConfig(group_size=128, int4_packing_format="tile_packed_to_4d", int4_choose_qparams_algorithm="hqq")
linear_config = AWQConfig(base_config, step="prepare")

# skip quantizing lm_head since it has different definition in vllm and transformers
quant_config = ModuleFqnToConfig({"_default": linear_config, "lm_head": None})

# your selected tasks, see https://github.com/EleutherAI/lm-evaluation-harness/blob/main/docs/new_task_guide.md for adding tasks to lm-eval
tasks = ["mmlu_abstract_algebra"]
calibration_limit = 10
max_seq_length = 2048

quantize_(
    model,
    quant_config,
)
tasks = []
TransformerEvalWrapper(
    model=model,
    tokenizer=tokenizer,
    max_seq_length=max_seq_length,
).run_eval(
    tasks=tasks,
    limit=calibration_limit,
)
linear_config = AWQConfig(base_config, step="convert")
quant_config = ModuleFqnToConfig({"_default": linear_config, "lm_head": None})
quantize_(model, quant_config)

quantized_model = model
linear_config = AWQConfig(base_config, step="prepare_for_loading")
quant_config = ModuleFqnToConfig({"_default": linear_config, "lm_head": None})
quantized_model.config.quantization_config = TorchAoConfig(quant_config)


# Push to hub
USER_ID = "YOUR_USER_ID"
MODEL_NAME = model_id.split("/")[-1]
save_to = f"{USER_ID}/{MODEL_NAME}-AWQ-INT4"
quantized_model.push_to_hub(save_to, safe_serialization=False)
tokenizer.push_to_hub(save_to)

# Manual Testing
prompt = "Hey, are you conscious? Can you talk to me?"
messages = [
    {
        "role": "system",
        "content": "",
    },
    {"role": "user", "content": prompt},
]
templated_prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
print("Prompt:", prompt)
print("Templated prompt:", templated_prompt)
inputs = tokenizer(
    templated_prompt,
    return_tensors="pt",
).to("cuda")
generated_ids = quantized_model.generate(**inputs, max_new_tokens=128)
output_text = tokenizer.batch_decode(
    generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print("Response:", output_text[0][len(prompt):])
```

Note: to `push_to_hub` you need to run
```Shell
pip install -U "huggingface_hub[cli]"
huggingface-cli login
```
and use a token with write access, from https://huggingface.co/settings/tokens

# Model Quality
We rely on [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) to evaluate the quality of the quantized model.

| Benchmark                        |                |                        |                           |
|----------------------------------|----------------|------------------------|---------------------------|
|                                  | Qwen/Qwen3-8B  | jerryzh168/Qwen3-8B-INT4-skip_lm_head  | pytorch/Qwen3-8B-AWQ-INT4 |
| mmlu_abstract_algebra            | 58             |  55                  |           56                |

Note that we only calibrate on a single `mmlu_abstract_algebra` task instead of the entire `mmlu` task since `mmlu` contains many different types of tasks and calibrating
on all of them does not necessarily improve the accuracy for all the tasks, since it's harder to faithfully represent the distribution of data from all types of tasks with a selected
small calibration sample data.

Note: we skipped quantization for `lm_head` because in transformers lm_head is a `Linear` but in vllm lm_head becomes [ParallelLMHead](https://github.com/vllm-project/vllm/blob/3e903b6cb4292ca1425a37cb809c1e3cddfdadcb/vllm/model_executor/models/qwen3.py#L294)
and the linear weight no longer works there.


<details>
<summary> Reproduce Model Quality Results </summary>

Need to install lm-eval from source:
https://github.com/EleutherAI/lm-evaluation-harness#install

## baseline
```Shell
lm_eval --model hf --model_args pretrained=Qwen/Qwen3-8B --tasks bbh --device cuda:0 --batch_size 8
```

## AWQ-INT4
```Shell
export MODEL=pytorch/Qwen3-8B-AWQ-INT4
lm_eval --model hf --model_args pretrained=$MODEL --tasks bbh --device cuda:0 --batch_size 8
```
</details>




# Peak Memory Usage

## Results

| Benchmark        |                |                                |                                |
|------------------|----------------|--------------------------------|--------------------------------|
|                  | Qwen/Qwen3-8B  | jerryzh168/Qwen3-8B-INT4-skip_lm_head |  pytorch/Qwen3-8B-AWQ-INT4     |
| Peak Memory (GB) | 16.47          | 7.82 (53% reduction)           |          7.82 (53% reduction)  |



<details>
<summary> Reproduce Peak Memory Usage Results </summary>

We can use the following code to get a sense of peak memory usage during inference:

```Py
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TorchAoConfig

# use "Qwen/Qwen3-8B" or "pytorch/Qwen3-8B-AWQ-INT4"
model_id = "pytorch/Qwen3-8B-AWQ-INT4"
quantized_model = AutoModelForCausalLM.from_pretrained(model_id, device_map="cuda:0", torch_dtype=torch.bfloat16)
tokenizer = AutoTokenizer.from_pretrained(model_id)

torch.cuda.reset_peak_memory_stats()

prompt = "Hey, are you conscious? Can you talk to me?"
messages = [
    {
        "role": "system",
        "content": "",
    },
    {"role": "user", "content": prompt},
]
templated_prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
print("Prompt:", prompt)
print("Templated prompt:", templated_prompt)
inputs = tokenizer(
    templated_prompt,
    return_tensors="pt",
).to("cuda")
generated_ids = quantized_model.generate(**inputs, max_new_tokens=128)
output_text = tokenizer.batch_decode(
    generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print("Response:", output_text[0][len(prompt):])

mem = torch.cuda.max_memory_reserved() / 1e9
print(f"Peak Memory Usage: {mem:.02f} GB")
```

</details>




# Model Performance

## Results (H100 machine)
| Benchmark (Latency)              |                |                           |                           |
|----------------------------------|----------------|---------------------------|---------------------------|
|                                  | Qwen/Qwen3-8B  | jerryzh168/Qwen3-8B-INT4-skip_lm_head  | pytorch/Qwen3-8B-AWQ-INT4 |
| latency (batch_size=1)           | 2.46s          | 1.40s (1.76x speedup)     | 1.83s (1.34x speedup)     |

<details>
<summary> Reproduce Model Performance Results </summary>

## Setup

Get vllm source code:
```Shell
git clone [email protected]:vllm-project/vllm.git
```

Install vllm
```
VLLM_USE_PRECOMPILED=1 pip install --editable .
```

Run the benchmarks under `vllm` root folder:

## benchmark_latency

### baseline
```Shell
export MODEL=Qwen/Qwen3-8B
python benchmarks/benchmark_latency.py --input-len 256 --output-len 256 --model $MODEL --batch-size 1
```

### AWQ-INT4
```Shell
export MODEL=pytorch/Qwen3-8B-AWQ-INT4
VLLM_DISABLE_COMPILE_CACHE=1 python benchmarks/benchmark_latency.py --input-len 256 --output-len 256 --model $MODEL --batch-size 1
```

## benchmark_serving

We benchmarked the throughput in a serving environment.

Download sharegpt dataset:

```Shell
wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json
```



Other datasets can be found in: https://github.com/vllm-project/vllm/tree/main/benchmarks

Note: you can change the number of prompts to be benchmarked with `--num-prompts` argument for `benchmark_serving` script.

### baseline
Server:
```Shell
export MODEL=Qwen/Qwen3-8B
vllm serve $MODEL --tokenizer $MODEL -O3
```

Client:
```Shell
export MODEL=Qwen/Qwen3-8B
python benchmarks/benchmark_serving.py --backend vllm --dataset-name sharegpt --tokenizer $MODEL --dataset-path ./ShareGPT_V3_unfiltered_cleaned_split.json --model $MODEL --num-prompts 1
```

### AWQ-INT4
Server:
```Shell
export MODEL=pytorch/Qwen3-8B-AWQ-INT4
VLLM_DISABLE_COMPILE_CACHE=1 vllm serve $MODEL --tokenizer $MODEL -O3 --pt-load-map-location cuda:0
```

Client:
```Shell
export MODEL=pytorch/Qwen3-8B-AWQ-INT4
python benchmarks/benchmark_serving.py --backend vllm --dataset-name sharegpt --tokenizer $MODEL --dataset-path ./ShareGPT_V3_unfiltered_cleaned_split.json --model $MODEL --num-prompts 1
```
</details>




# Paper: TorchAO: PyTorch-Native Training-to-Serving Model Optimization
The model's quantization is powered by **TorchAO**, a framework presented in the paper [TorchAO: PyTorch-Native Training-to-Serving Model Optimization](https://huggingface.co/papers/2507.16099).

**Abstract:** We present TorchAO, a PyTorch-native model optimization framework leveraging quantization and sparsity to provide an end-to-end, training-to-serving workflow for AI models. TorchAO supports a variety of popular model optimization techniques, including FP8 quantized training, quantization-aware training (QAT), post-training quantization (PTQ), and 2:4 sparsity, and leverages a novel tensor subclass abstraction to represent a variety of widely-used, backend agnostic low precision data types, including INT4, INT8, FP8, MXFP4, MXFP6, and MXFP8. TorchAO integrates closely with the broader ecosystem at each step of the model optimization pipeline, from pre-training (TorchTitan) to fine-tuning (TorchTune, Axolotl) to serving (HuggingFace, vLLM, SGLang, ExecuTorch), connecting an otherwise fragmented space in a single, unified workflow. TorchAO has enabled recent launches of the quantized Llama 3.2 1B/3B and LlamaGuard3-8B models and is open-source at this https URL .

# Resources
*   **Official TorchAO GitHub Repository:** [https://github.com/pytorch/ao](https://github.com/pytorch/ao)
*   **TorchAO Documentation:** [https://docs.pytorch.org/ao/stable/index.html](https://docs.pytorch.org/ao/stable/index.html)


# Disclaimer
PyTorch has not performed safety evaluations or red teamed the quantized models. Performance characteristics, outputs, and behaviors may differ from the original models. Users are solely responsible for selecting appropriate use cases, evaluating and mitigating for accuracy, safety, and fairness, ensuring security, and complying with all applicable laws and regulations.

Nothing contained in this Model Card should be interpreted as or deemed a restriction or modification to the licenses the models are released under, including any limitations of liability or disclaimers of warranties provided therein.