--- library_name: transformers tags: - torchao - qwen - qwen3 - nlp - code - math - chat - conversational license: mit language: - multilingual base_model: - Qwen/Qwen3-4B pipeline_tag: text-generation --- [Qwen3-4B](https://huggingface.co/Qwen/Qwen3-4B) is quantized by the PyTorch team using [torchao](https://huggingface.co/docs/transformers/main/en/quantization/torchao) with 8-bit embeddings and 8-bit dynamic activations with 4-bit weight linears (8da4w). The model is suitable for mobile deployment with [ExecuTorch](https://github.com/pytorch/executorch). We provide the [quantized pte](https://huggingface.co/pytorch/Qwen3-4B-8da4w/blob/main/qwen3-4B-8da4w-1024-cxt.pte) for direct use in ExecuTorch. (The provided pte file is exported with a max_seq_length/max_context_length of 1024; if you wish to change this, re-export the quantized model following the instructions in [Exporting to ExecuTorch](#exporting-to-executorch).) # Running in a mobile app The [pte file](https://huggingface.co/pytorch/Qwen3-4B-8da4w/blob/main/qwen3-4B-8da4w-1024-cxt.pte) can be run with ExecuTorch on a mobile phone. See the [instructions](https://pytorch.org/executorch/main/llm/llama-demo-ios.html) for doing this in iOS. On iPhone 15 Pro, the model runs at 14.8 tokens/sec and uses 3379 Mb of memory. ![image/png](https://cdn-uploads.huggingface.co/production/uploads/66049fc71116cebd1d3bdcf4/eVHB7fVllmwVauKJvGu0d.png) # Quantization Recipe First need to install the required packages: ```Shell pip install git+https://github.com/huggingface/transformers@main pip install git+https://github.com/pytorch/ao.git@main ``` ## Untie Embedding Weights We want to quantize the embedding and lm_head differently. Since those layers are tied, we first need to untie the model: ```Py from transformers import ( AutoModelForCausalLM, AutoProcessor, AutoTokenizer, ) import torch model_id = "Qwen/Qwen3-4B" untied_model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto", device_map="auto") tokenizer = AutoTokenizer.from_pretrained(model_id) print(untied_model) from transformers.modeling_utils import find_tied_parameters print("tied weights:", find_tied_parameters(untied_model)) if getattr(untied_model.config.get_text_config(decoder=True), "tie_word_embeddings"): setattr(untied_model.config.get_text_config(decoder=True), "tie_word_embeddings", False) untied_model._tied_weights_keys = [] untied_model.lm_head.weight = torch.nn.Parameter(untied_model.lm_head.weight.clone()) print("tied weights:", find_tied_parameters(untied_model)) USER_ID = "YOUR_USER_ID" MODEL_NAME = model_id.split("/")[-1] save_to = f"{USER_ID}/{MODEL_NAME}-untied-weights" untied_model.push_to_hub(save_to) tokenizer.push_to_hub(save_to) # or save locally save_to_local_path = f"{MODEL_NAME}-untied-weights" untied_model.save_pretrained(save_to_local_path) tokenizer.save_pretrained(save_to) ``` 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 ## Quantization We used following code to get the quantized model: ```Py from transformers import ( AutoModelForCausalLM, AutoProcessor, AutoTokenizer, TorchAoConfig, ) from torchao.quantization.quant_api import ( IntxWeightOnlyConfig, Int8DynamicActivationIntxWeightConfig, ModuleFqnToConfig, quantize_, ) from torchao.quantization.granularity import PerGroup, PerAxis import torch # we start from the model with untied weights model_id = "Qwen/Qwen3-4B" USER_ID = "YOUR_USER_ID" MODEL_NAME = model_id.split("/")[-1] untied_model_id = f"{USER_ID}/{MODEL_NAME}-untied-weights" untied_model_local_path = f"{MODEL_NAME}-untied-weights" embedding_config = IntxWeightOnlyConfig( weight_dtype=torch.int8, granularity=PerAxis(0), ) linear_config = Int8DynamicActivationIntxWeightConfig( weight_dtype=torch.int4, weight_granularity=PerGroup(32), weight_scale_dtype=torch.bfloat16, ) quant_config = ModuleFqnToConfig({"_default": linear_config, "model.embed_tokens": embedding_config}) quantization_config = TorchAoConfig(quant_type=quant_config, include_embedding=True, untie_embedding_weights=True, modules_to_not_convert=[]) # either use `untied_model_id` or `untied_model_local_path` quantized_model = AutoModelForCausalLM.from_pretrained(untied_model_id, torch_dtype=torch.float32, device_map="auto", quantization_config=quantization_config) tokenizer = AutoTokenizer.from_pretrained(model_id) # Push to hub MODEL_NAME = model_id.split("/")[-1] save_to = f"{USER_ID}/{MODEL_NAME}-8da4w" 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):]) ``` The response from the manual testing is: ``` Hello! I'm Qwen, a large language model developed by Alibaba Cloud. While I don't have consciousness or personal experiences, I can engage in conversations with you and help answer questions. I can talk to you, share thoughts, and even have fun! What's on your mind? ``` # Model Quality | Benchmark | | | |----------------------------------|----------------|---------------------------| | | Qwen3-4B | Qwen3-4B-8da4w | | **Popular aggregated benchmark** | | | | mmlu | 68.38 | 66.74 | | mmlu_pro | 49.71 | 46.73 | | bbh | 74.86 | 67.47 | | **Reasoning** | | | | gpqa_main_zeroshot | 33.93 | 31.03 | | **Multilingual** | | | | m_mmlu | 50.41 | 47.13 | | mgsm_en_cot_en | 30.40 | 29.20 | | **Math** | | | | gsm8k | 84.76 | 82.87 | | leaderboard_math_hard (v3) | 48.19 | 44.94 | | **Overall** | 55.08 | 52.01 |
Reproduce Model Quality Results We rely on [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) to evaluate the quality of the quantized model. Need to install lm-eval from source: https://github.com/EleutherAI/lm-evaluation-harness#install ## baseline ```Shell lm_eval --model hf --model_args pretrained=Qwen3/Qwen3-4B --tasks mmlu --device cuda:0 --batch_size auto ``` ## int8 dynamic activation and int4 weight quantization (8da4w) ```Shell lm_eval --model hf --model_args pretrained=pytorch/Qwen3-4B-8da4w --tasks mmlu --device cuda:0 --batch_size auto ```
# Exporting to ExecuTorch We can run the quantized model on a mobile phone using [ExecuTorch](https://github.com/pytorch/executorch). Once ExecuTorch is [set-up](https://pytorch.org/executorch/main/getting-started.html), exporting and running the model on device is a breeze. We first convert the [quantized checkpoint](https://huggingface.co/pytorch/Qwen3-4B-8da4w/blob/main/pytorch_model.bin) to one ExecuTorch's LLM export script expects by renaming some of the checkpoint keys. The following script does this for you. We have uploaded the converted checkpoint [pytorch_model_converted.bin](https://huggingface.co/pytorch/Qwen3-4B-8da4w/blob/main/pytorch_model_converted.bin) for convenience. ```Shell python -m executorch.examples.models.qwen3.convert_weights $(huggingface-cli download pytorch/Qwen3-4B-8da4w) pytorch_model_converted.bin ``` Once the checkpoint is converted, we can export to ExecuTorch's pte format with the XNNPACK delegate. The below command exports with a max_seq_length/max_context_length of 1024, but it can be changed as desired. ```Shell PARAMS="executorch/examples/models/qwen3/4b_config.json" python -m executorch.examples.models.llama.export_llama \ --model "qwen3-4b" \ --checkpoint "pytorch_model_converted.bin" \ --params "$PARAMS" \ -kv \ --use_sdpa_with_kv_cache \ -d fp32 -X \ --metadata '{"get_bos_id":199999, "get_eos_ids":[200020,199999]}' \ --max_seq_length 1024 \ --max_context_length 1024 \ --output_name="qwen3-4b-8da4w-1024-cxt.pte" ``` After that you can run the model in a mobile app (see [Running in a mobile app](#running-in-a-mobile-app)). # 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.