# Avoid loading checkpoint shards for each inference
**URL:**
**Author:**  [@alzaia](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Falzaia)%5C
**Post date:** [December 19, 2023, 3:35pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Favoid-loading-checkpoint-shards-for-each-inference%2F66357%2F1 "2023-12-19T15:35:33Z")
I am currently using LLaVA for inference and I was wondering if there was a way to avoid reloading the checkpoint shards every time I predict a new sample. For reference, I am closely following the doc and the colab here: [llava-hf/llava-1.5-7b-hf · Hugging Face](https://huggingface.co/llava-hf/llava-1.5-7b-hf).
---
**Author:**  [@enochlev](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fenochlev)%5C
**Post date:** [December 19, 2023, 6:31pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Favoid-loading-checkpoint-shards-for-each-inference%2F66357%2F2 "2023-12-19T18:31:22Z")
provide us with a snippet from your code
otherwize I can only say don’t call this function every time you do an inference
```auto
LlavaForConditionalGeneration.from_pretrained
```
---
**Author:**  [@alzaia](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Falzaia)%5C
**Post date:** [December 19, 2023, 7:48pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Favoid-loading-checkpoint-shards-for-each-inference%2F66357%2F3 "2023-12-19T19:48:56Z")
I’m using the pipeline from HuggingFace, so my code is super simple:
```auto
image = Image.open(image_path)
pipe = pipeline(
"image-to-text",
model=model_id,
model_kwargs={"quantization_config": quantization_config},
)
output = pipe(
image, prompt=text_prompt, generate_kwargs={"max_new_tokens": 200}
)
```
Should I use pure transformers library instead to have more control?