--- license: apache-2.0 language: - en pipeline_tag: text-generation library_name: transformers tags: - nlp - llm - mllm --- # CrystalChat-7B-Web2Code: a fully-reproducible vision large language model based on CrystalChat-7B LLM for webpage code generation ## Model Description CrystalChat-7B based multi-modal large language model (MLLM) mimics the training recipe used for Vicuna-7B based [LLaVa-v1.5](https://huggingface.co./docs/transformers/main/model_doc/llava). CrystalChat-7B based MLLMs models are entirely transparent, having open-sourced all materials, including code, data, model checkpoint, intermediate results, and more at [TODO: Add paper link](). CrystalChat-7B-Web2Code MLLM is specialized in webpage images-to-html code generation. ### About CrystalChat-7B-Web2Code: * 7 billion parameter LLM * CLIP ViT-L/14-336px vision encoder * Languages: English * Models Released: CrystalChat-7B-Web2Code * Trained in 2 stages * License: ? Crystal-based models were developed as a collaboration between [MBZUAI](https://mbzuai.ac.ae/institute-of-foundation-models/), [Petuum](https://www.petuum.com/), and [LLM360](https://www.llm360.ai/)????. ## Evaluation General Evaluation Metrics for MLLMs. MME serves as an extensive evaluative benchmark, aiming to assess perceptual and cognitive capability of MLLMs within 14 sub-tasks. Additionally, we also evaluate the performance of our models on text-oriented visual question answering tasks employing a diverse set of benchmark datasets including ScienceQA and TextVQA. Furthermore, we assess our models’ ability toward anti-hallucination through POPE. | LLM Backbone | MME-P | MME-C | POPE | SciQA | TextVQA | |-----------------------------------|---------|--------|-------|--------|---------| | CrystalCoder-7B | 1359.83 | 238.92 | 86.182 | 64.15 | 50.39 | | CrystalChat-7B | 1456.53 | **308.21** | 86.96 | 67.77 | **57.84** | | Vicuna-7B | **1481.12** | 302.85 | **87.174** | **67.97** | 56.49 | *Table 1: Comparison of different LLM backbones on visual language understanding benchmarks. All models are instruction-tuned on the general domain data (i.e. LLaVA)* TODO: Add general and code evaluations once jason confirms ## Data and Training Details ### Pretrain Data LLaVA Visual Instruct Pretrain LCS-558K is a filtered subset of the LAION, CC, and SBU datasets, featuring a more balanced distribution of concept coverage. The file includes multimodal synthesized conversations generated from image-caption pairs by incorporating randomly selected instructions such as "Describe this image." It is used for pretraining in LLaVA, with the raw CC-3M caption serving as the default answer. ### Finetune Data The finetuning data contains the following: #### LLaVa Finetuning Data The dataset chosen was created by LLaVA with academic-task-oriented VQA data mixture and data from ShareGPT. LLaVA Visual Instruct 150K is a dataset of GPT-generated multimodal instruction-following data. It is designed for visual instruction tuning and aims to develop large multimodal models with capabilities akin to GPT-4 in both vision and language. | Data | Size | Response formatting prompts | |---------------|------|--------------------------------------------------------------------------| | LLaVA [36] | 158K | – | | ShareGPT [46] | 40K | – | | VQAv2 [19] | 83K | Answer the question using a single word or phrase. | | GQA [21] | 72K | Answer the question using a single word or phrase. | | OKVQA [41] | 9K | Answer the question using a single word or phrase. | | OCRVQA [42] | 80K | Answer the question using a single word or phrase. | | A-OKVQA [45] | 66K | Answer with the option’s letter from the given choices directly. | | TextCaps [47] | 22K | Provide a one-sentence caption for the provided image. | | RefCOCO [24, 40] | 48K | Note: randomly choose between the two formats. Provide a short description for this region. | | VG [25] | 86K | Provide the bounding box coordinate of the region this sentence describes. | | **Total** | **665K** | | *Table 2. Instruction-following Data Mixture of LLaVA-1.5.* #### Web2Code Data The Web2Code instruction tuning dataset was released in [Web2Code: A Large-scale Webpage-to-Code Dataset and Evaluation Framework for Multimodal LLMs](TODO: Add link). The dataset construction and instruction generation process involves four key components: DWCG: We created new webpage image-code pair data DWCG by generating high-quality HTML webpage-code pairs following the CodeAlpaca prompt using GPT-3.5 and converting them into instruction-following data. DWCGR: We refined existing webpage code generation data by transforming existing datasets, including WebSight and Pix2Code, into an instruction-following data format similar to LLaVA data, so they can be used as instruction-following data to train MLLMs. DWU: We created new text question-answer pair data by generating a new question-answer pair dataset utilizing our new GPT-3.5 generated data for webpage understanding. DWUR: We refined the WebSRC question-answer data to improve its quality using GPT-4. ### Code Datasets | Dataset | DWCG (ours) | DWCGR (ours) | |---------|-------------|-------------------| | **Instruction** | ✓ | ✓ | | **Source** | Synthetic | Synthetic | | **Size** | 60K | 824.7K | | **Avg Length (tokens)** | 471.8±162.3 | 652.85±157.0 | | **Avg Tag Count** | 28.1±10.6 | 35.3±9.0 | | **Avg DOM Depth** | 5.3±1.0 | 6.5±1.0 | | **Avg Unique Tags** | 13.6±2.7 | 13.5±2.5 | *Table 3. DWCG is a newly generated GPT-3.5-based dataset, while DWCGR is the refined dataset that utilizes WebSight and Pix2Code datasets* ### Webpage Understanding Datasets | Dataset | DWU | DWUR | |---------------|---------|-----------------| | **Instruction** | ✓ | ✓ | | **Size** | 243.5K | 51.5K | *Table 4. Distribution of DWU and DWUR datasets. Both datasets include high-quality question-answer pairs for webpage understanding.* #TODO: check if this is needed, if yes, replace with corresponding for code model ## Stage 2 - Finetuning | Checkpoints | | | ----------- | ----------- | | [CrystalChat](https://huggingface.co./qazimbhat1/my-model-repo3/tree/main) | | [CrystalCoder](https://huggingface.co./qazimbhat1/Crystal-based-MLLM-7B/tree/Crystal-coder-7B) | ## Stage 1 - Pretraining | Checkpoints | | | ----------- | ----------- | | [CrystalChat](https://huggingface.co./qazimbhat1/Crystal-based-MLLM-7B/tree/Crystal-based-MLLM-7B-pretrain) | | [CrystalCoder](https://huggingface.co./qazimbhat1/Crystal-based-MLLM-7B/tree/Crystal-coder-7B-pretrain) | [to find all branches: git branch -a] ## Examples TODO: Add image as sample example Example 1:
Original Input image
*Image 1. Original Input Image.*
CrsytalChat-7B model generated output
*Image 2. CrystalChat-7B-Web2Code model output.* Example 2:
CrsytalChat-7B model generated output
*Image 3. Hand-drawn webpage input to CrystalChat-7B-Web2Code generated output.* ## Loading Crystal ```python from transformers import AutoModelForCausalLM, AutoTokenizer tokenizer = AutoTokenizer.from_pretrained( "LLM360/CrystalChat-7B-MLLM", padding_side="right", trust_remote_code=True) model = AutoModelForCausalLM.from_pretrained( "LLM360/CrystalChat-7B-MLLM", trust_remote_code=True, torch_dtype=torch.float16, device_map='auto', low_cpu_mem_usage=True ) ``` ## LLM-360 LLM-360 is an open research lab enabling community-owned AGI through open-source large model research and development. Crystal-based Models enables community-owned AGI by creating standards and tools to advance the bleeding edge of LLM capability and empower knowledge transfer, research, and development. We believe in a future where artificial general intelligence (AGI) is created by the community, for the community. Through an open ecosystem of equitable computational resources, high-quality data, and flowing technical knowledge, we can ensure ethical AGI development and universal access for all innovators. [Visit us](https://www.llm360.ai/) ## Citation **BibTeX:** ```bibtex @article{ title={}, author={}, year={}, } ```