--- language: - en pipeline_tag: text-generation license: llama2 --- # gemma-2-9b-it-quantized.w4a16 ## Model Overview - **Model Architecture:** Gemma-2 - **Input:** Text - **Output:** Text - **Model Optimizations:** - **Weight quantization:** INT4 - **Intended Use Cases:** Intended for commercial and research use in English. Similarly to [gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it), this models is intended for assistant-like chat. - **Out-of-scope:** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in languages other than English. - **Release Date:** 8/16/2024 - **Version:** 1.0 - **License(s)**: [Gemma](https://ai.google.dev/gemma/terms) - **Model Developers:** Neural Magic Quantized version of [gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it). It achieves an average score of 73.62 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 73.23. ### Model Optimizations This model was obtained by quantizing the weights of [gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it) to INT4 data type. This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 75%. Only the weights of the linear operators within transformers blocks are quantized. Symmetric group-wise quantization is applied, in which a linear scaling per group maps the INT4 and floating point representations of the quantized weights. The [GPTQ](https://arxiv.org/abs/2210.17323) algorithm is applied for quantization, as implemented in the [llm-compressor](https://github.com/vllm-project/llm-compressor) library. Quantization is performed with 1% damping factor, group-size as 128 and 512 sequences sampled from [Open-Platypus](https://huggingface.co/datasets/garage-bAInd/Open-Platypus). ## Deployment ### Use with vLLM This model can be deployed efficiently using the [vLLM](https://docs.vllm.ai/en/latest/) backend, as shown in the example below. ```python from vllm import LLM, SamplingParams from transformers import AutoTokenizer model_id = "neuralmagic/gemma-2-9b-it-quantized.w4a16" sampling_params = SamplingParams(temperature=0.6, top_p=0.9, max_tokens=256) tokenizer = AutoTokenizer.from_pretrained(model_id) messages = [ {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"}, {"role": "user", "content": "Who are you? Please respond in pirate speak."}, ] prompts = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) llm = LLM(model=model_id, tensor_parallel_size=2) outputs = llm.generate(prompts, sampling_params) generated_text = outputs[0].outputs[0].text print(generated_text) ``` vLLM aslo supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details. ### Use with transformers The following example contemplates how the model can be deployed in Transformers using the `generate()` function. ```python from transformers import AutoTokenizer, AutoModelForCausalLM model_id = "neuralmagic/gemma-2-9b-it-quantized.w4a16" tokenizer = AutoTokenizer.from_pretrained(model_id) model = AutoModelForCausalLM.from_pretrained( model_id, torch_dtype="auto", device_map="auto", ) messages = [ {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"}, {"role": "user", "content": "Who are you? Please respond in pirate speak"}, ] input_ids = tokenizer.apply_chat_template( messages, add_generation_prompt=True, return_tensors="pt" ).to(model.device) terminators = [ tokenizer.eos_token_id, tokenizer.convert_tokens_to_ids("<|eot_id|>") ] outputs = model.generate( input_ids, max_new_tokens=256, eos_token_id=terminators, do_sample=True, temperature=0.6, top_p=0.9, ) response = outputs[0][input_ids.shape[-1]:] print(tokenizer.decode(response, skip_special_tokens=True)) ``` ## Creation This model was created by using the [llm-compressor](https://github.com/vllm-project/llm-compressor) library as presented in the code snipet below. ```python from transformers import AutoTokenizer from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot from llmcompressor.modifiers.quantization import GPTQModifier from datasets import load_dataset import random model_id = "google/gemma-2-9b-it" num_samples = 512 max_seq_len = 4096 tokenizer = AutoTokenizer.from_pretrained(model_id) preprocess_fn = lambda example: {"text": "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n{text}".format_map(example)} dataset_name = "neuralmagic/LLM_compression_calibration" dataset = load_dataset(dataset_name, split="train") ds = dataset.shuffle().select(range(num_samples)) ds = ds.map(preprocess_fn) examples = [ tokenizer( example["text"], padding=False, max_length=max_seq_len, truncation=True, ) for example in ds ] recipe = GPTQModifier( targets="Linear", scheme="W4A16", ignore=["lm_head"], dampening_frac=0.01, ) model = SparseAutoModelForCausalLM.from_pretrained( model_id, device_map="auto", trust_remote_code=True, ) oneshot( model=model, dataset=ds, recipe=recipe, max_seq_length=max_seq_len, num_calibration_samples=num_samples, ) model.save_pretrained("gemma-2-9b-it-quantized.w4a16") ``` ## Evaluation The model was evaluated on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) leaderboard tasks (version 1) with the [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness/tree/383bbd54bc621086e05aa1b030d8d4d5635b25e6) (commit 383bbd54bc621086e05aa1b030d8d4d5635b25e6) and the [vLLM](https://docs.vllm.ai/en/stable/) engine, using the following command: ``` lm_eval \ --model vllm \ --model_args pretrained="neuralmagic/gemma-2-9b-it-quantized.w4a16",dtype=auto,tensor_parallel_size=2,gpu_memory_utilization=0.4,add_bos_token=True,max_model_len=4096,trust_remote_code=True \ --tasks openllm \ --batch_size auto ``` ### Accuracy #### Open LLM Leaderboard evaluation scores
Benchmark gemma-2-9b-it gemma-2-9b-it-quantized.w4a16(this model) Recovery
MMLU (5-shot) 72.28 71.36 98.72%
ARC Challenge (25-shot) 71.5 70.98 99.27%
GSM-8K (5-shot, strict-match) 76.26 79.83 104.68%
Hellaswag (10-shot) 81.91 81.29 99.24%
Winogrande (5-shot) 77.11 78.29 101.53%
TruthfulQA (0-shot) 60.32 59.97 99.42%
Average 73.23 73.62 100.53%