alexmarques's picture
Update README.md
a241b9c verified
---
pipeline_tag: text-generation
tags:
- int8
- vllm
license: gemma
---
# gemma-2-2b-it-quantized.w8a16
## Model Overview
- **Model Architecture:** Gemma 2
- **Input:** Text
- **Output:** Text
- **Model Optimizations:**
- **Weight quantization:** INT8
- **Intended Use Cases:** Intended for commercial and research use in English. Similarly to [gemma-2-2b](https://huggingface.co/google/gemma-2-2b), this is a pre-trained base model that can be used as is or specialized to specific domains.
- **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:** 8/13/2024
- **Version:** 1.0
- **License(s):** [gemma](https://ai.google.dev/gemma/terms)
- **Model Developers:** Neural Magic
Quantized version of [gemma-2-2b](https://huggingface.co/google/gemma-2-2b).
It achieves an average score of 52.03 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 52.16.
### Model Optimizations
This model was obtained by quantizing the weights of [gemma-2-2b](https://huggingface.co/google/gemma-2-2b) to INT8 data type.
This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%.
Only the weights of the linear operators within transformers blocks are quantized. Symmetric per-channel quantization is applied, in which a linear scaling per output dimension maps the INT8 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.
GPTQ used a 1% damping factor and 256 sequences sequences taken from Neural Magic's [LLM compression calibration dataset](https://huggingface.co/datasets/neuralmagic/LLM_compression_calibration).
## 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-2-2b-quantized.w8a16"
sampling_params = SamplingParams(temperature=0.6, top_p=0.9, max_tokens=256)
tokenizer = AutoTokenizer.from_pretrained(model_id)
messages = [
{"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)
outputs = llm.generate(prompts, sampling_params)
generated_text = outputs[0].outputs[0].text
print(generated_text)
```
vLLM also supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details.
## 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 datasets import load_dataset
from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot
from llmcompressor.modifiers.quantization import GPTQModifier
model_id = "google/gemma-2-2b"
num_samples = 256
max_seq_len = 8192
tokenizer = AutoTokenizer.from_pretrained(model_id)
def preprocess_fn(example):
return {"text": tokenizer.apply_chat_template(example["messages"], add_generation_prompt=False, tokenize=False)}
ds = load_dataset("neuralmagic/LLM_compression_calibration", split="train")
ds = ds.shuffle().select(range(num_samples))
ds = ds.map(preprocess_fn)
recipe = GPTQModifier(
targets="Linear",
scheme="W8A16",
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-2-2b-quantized.w8a16")
```
## 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-2-2b-quantized.w8a16",dtype=auto,gpu_memory_utilization=0.4,add_bos_token=True,max_model_len=4096 \
--tasks openllm \
--batch_size auto
```
### Accuracy
#### Open LLM Leaderboard evaluation scores
<table>
<tr>
<td><strong>Benchmark</strong>
</td>
<td><strong>gemma-2-2b</strong>
</td>
<td><strong>gemma-2-2b-quantized.w8a16 (this model)</strong>
</td>
<td><strong>Recovery</strong>
</td>
</tr>
<tr>
<td>MMLU (5-shot)
</td>
<td>53.01
</td>
<td>52.86
</td>
<td>99.7%
</td>
</tr>
<tr>
<td>ARC Challenge (25-shot)
</td>
<td>53.92
</td>
<td>54.27
</td>
<td>100.6%
</td>
</tr>
<tr>
<td>GSM-8K (5-shot, strict-match)
</td>
<td>24.03
</td>
<td>23.81
</td>
<td>99.1%
</td>
</tr>
<tr>
<td>Hellaswag (10-shot)
</td>
<td>74.74
</td>
<td>74.56
</td>
<td>99.8%
</td>
</tr>
<tr>
<td>Winogrande (5-shot)
</td>
<td>70.96
</td>
<td>70.56
</td>
<td>99.4%
</td>
</tr>
<tr>
<td>TruthfulQA (0-shot)
</td>
<td>36.29
</td>
<td>36.15
</td>
<td>99.6%
</td>
</tr>
<tr>
<td><strong>Average</strong>
</td>
<td><strong>52.16</strong>
</td>
<td><strong>52.03</strong>
</td>
<td><strong>99.8%</strong>
</td>
</tr>
</table>