Downloads · 30 days
43
8% of all-time downloads
SLIMER-LLaMA3
expertai/SLIMER-LLaMA3
SLIMER-LLaMA3 is a text generation model from expertai. Use it when you need the model to write or continue text. The card lists the license as llama3.1.
This LLaMA-3 based model scores +12.5 % wrt paper's original SLIMER LLaMA-2.
All-time downloads
506
Public
Parameters
8B
16.1 GB on disk
Likes
3
Public
Repo makeup
Click a slice to open those files.
.safetensors16.1 GB · 100%
At a glance
- Task
- Text Generation
- License
- llama3.1
- Model type
- llama
- Access
- Public
- Created
- Oct 8, 2024
- Updated
- Nov 5, 2024
- SHA
- 639cb912
Try a prompt
Base models
Datasets
Model card
From the Hugging Face model README
SLIMER-3: Show Less Instruct More Entity Recognition LLaMA3
This LLaMA-3 based model scores +12.5 % wrt paper's original SLIMER LLaMA-2.
⚠️ A more powerful model, which scores +17% and allows parallel NEs extraction, can be found at: https://huggingface.co/expertai/SLIMER-PARALLEL-LLaMA3
GitHub repository: https://github.com/andrewzamai/SLIMER
SLIMER is an LLM specifically instructed for zero-shot NER on English language.
SLIMER for Italian language can be found at: https://huggingface.co/expertai/LLaMAntino-3-SLIMER-IT
Instructed on a reduced number of samples, it is designed to tackle never-seen-before Named Entity tags by leveraging a prompt enriched with a DEFINITION and GUIDELINES for the NE to be extracted.
<!DOCTYPE html> <html> <head> <title>Instruction Tuning Prompt</title> <style> .container { border: none; padding: 5px; width: 300px; margin: 0 auto; font-family: Arial, sans-serif; font-size: 8px; border-radius: 10px; /* Rounded borders for container */ overflow: hidden; /* Ensure child elements respect container's rounded borders */ background-color: #f0f0f0 } .header { background-color: black; color: white; padding: 5px; text-align: center; font-weight: bold; font-size: 14px; border-top-left-radius: 10px; /* Rounded top-left corner */ border-top-right-radius: 10px; /* Rounded top-right corner */ } .content { padding: 5px; } .definition, .guidelines { padding: 5px; border-radius: 10px; /* Rounded borders for definition and guidelines */ } .definition { background-color: #ffc773; } .guidelines { background-color: #73d7ff; } .footer { background-color: black; color: white; padding: 10px; font-weight: bold; border-bottom-left-radius: 10px; border-bottom-right-radius: 10px; } </style> </head> <body> <div class="container"> <div class="header">Instruction Tuning Prompt</div> <div class="content"> <p><b><|start_header_id|>user<|end_header_id|></b></p> <p>You are given a text chunk (delimited by triple quotes) and an instruction.<br> Read the text and answer to the instruction in the end.</p> <p>"""<br> {input text}<br> """</p> <p><b>Instruction:</b> Extract the Named Entities of type <b>DATE</b> from the text chunk you have read.</p> <p>You are given a <b>DEFINITION</b> and some <b>GUIDELINES</b>.</p> <div class="definition"> <p><b>DEFINITION:</b> <b>DATE</b> refers to specific points in time, including days, months, years, and relative time expressions like 'Week 2'.</p> </div> <div class="guidelines"> <p><b>GUIDELINES:</b> Avoid labeling non-specific time references like 'recently' or 'soon'. Exercise caution with ambiguous terms like 'May' (month or verb) and 'Wednesday Adams' (person's name which includes a day of the week).</p> </div> <p>Return a JSON list of instances of this Named Entity type (for example [\"text_span_1\", \"text_span_2\"]. Return an empty list [] if no instances are present. Return only the JSON list, no further motivations or introduction to the answer.</p> <p><b><|eot_id|><|start_header_id|>assistant<|end_header_id|></b></p> </div> <div class="footer"></div> </div> </body> </html>Currently existing approaches fine-tune on an extensive number of entity classes (around 13K) and assess zero-shot NER capabilities on Out-Of-Distribution input domains. SLIMER performs comparably to these state-of-the-art models on OOD input domains, while being trained only a reduced number of samples and a set of NE tags that overlap in lesser degree with test sets.
We extend the standard zero-shot evaluations (CrossNER and MIT) with BUSTER, which is characterized by financial entities that are rather far from the more traditional tags observed by all models during training. An inverse trend can be observed, with SLIMER emerging as the most effective in dealing with these unseen labels, thanks to its lighter instruction tuning methodology and the use of definition and guidelines.
<!DOCTYPE html> <html> <head> <style> table { width: 100%; border-collapse: collapse; font-size: 12px; } th, td { border: 1px none; padding: 4px; text-align: center; } th { background-color: #f2f2f2; } .col-model { width: 10%; } .col-backbone { width: 15%; } .col-params { width: 10%; } .col-mit, .col-crossner, .col-buster, .col-avg { width: 7%; } </style> </head> <body> <table> <thead> <tr> <th class="col-model">Model</th> <th class="col-backbone">Backbone</th> <th class="col-params">#Params</th> <th class="col-mit" colspan="2">MIT</th> <th class="col-crossner" colspan="5">CrossNER</th> <th class="col-buster">BUSTER</th> <th class="col-avg">AVG</th> </tr> <tr> <th></th> <th></th> <th></th> <th class="col-mit">Movie</th> <th class="col-mit">Restaurant</th> <th class="col-crossner">AI</th> <th class="col-crossner">Literature</th> <th class="col-crossner">Music</th> <th class="col-crossner">Politics</th> <th class="col-crossner">Science</th> <th class="col-buster"></th> <th class="col-avg"></th> </tr> </thead> <tbody> <tr> <td class="col-model">ChatGPT</td> <td class="col-backbone">gpt-3.5-turbo</td> <td class="col-params">-</td> <td class="col-mit">5.3</td> <td class="col-mit">32.8</td> <td class="col-crossner">52.4</td> <td class="col-crossner">39.8</td> <td class="col-crossner">66.6</td> <td class="col-crossner">68.5</td> <td class="col-crossner">67.0</td> <td class="col-buster">-</td> <td class="col-avg">-</td> </tr> <tr> <td class="col-model">InstructUIE</td> <td class="col-backbone">Flan-T5-xxl</td> <td class="col-params">11B</td> <td class="col-mit">63.0</td> <td class="col-mit">21.0</td> <td class="col-crossner">49.0</td> <td class="col-crossner">47.2</td> <td class="col-crossner">53.2</td> <td class="col-crossner">48.2</td> <td class="col-crossner">49.3</td> <td class="col-buster">-</td> <td class="col-avg">-</td> </tr> <tr> <td class="col-model">UniNER-type</td> <td class="col-backbone">LLaMA-1</td> <td class="col-params">7B</td> <td class="col-mit">42.4</td> <td class="col-mit">31.7</td> <td class="col-crossner">53.5</td> <td class="col-crossner">59.4</td> <td class="col-crossner">65.0</td> <td class="col-crossner">60.8</td> <td class="col-crossner">61.1</td> <td class="col-buster">34.8</td> <td class="col-avg">51.1</td> </tr> <tr> <td class="col-model">GoLLIE</td> <td class="col-backbone">Code-LLaMA</td> <td class="col-params">7B</td> <td class="col-mit">63.0</td> <td class="col-mit">43.4</td> <td class="col-crossner">59.1</td> <td class="col-crossner">62.7</td> <td class="col-crossner">67.8</td> <td class="col-crossner">57.2</td> <td class="col-crossner">55.5</td> <td class="col-buster" style="color: red;">27.7</td> <td class="col-avg">54.6</td> </tr> <tr> <td class="col-model">GLiNER-L</td> <td class="col-backbone">DeBERTa-v3</td> <td class="col-params">0.3B</td> <td class="col-mit">57.2</td> <td class="col-mit">42.9</td> <td class="col-crossner">57.2</td> <td class="col-crossner">64.4</td> <td class="col-crossner">69.6</td> <td class="col-crossner">72.6</td> <td class="col-crossner">62.6</td> <td class="col-buster" style="color: red;">26.6</td> <td class="col-avg">56.6</td> </tr> <tr> <td class="col-model">GNER-T5</td> <td class="col-backbone">Flan-T5-xxl</td> <td class="col-params">11B</td> <td class="col-mit">62.5</td> <td class="col-mit">51.0</td> <td class="col-crossner">68.2</td> <td class="col-crossner">68.7</td> <td class="col-crossner">81.2</td> <td class="col-crossner">75.1</td> <td class="col-crossner">76.7</td> <td class="col-buster" style="color: red;">27.9</td> <td class="col-avg">63.9</td> </tr> <tr> <td class="col-model">GNER-LLaMA</td> <td class="col-backbone">LLaMA-1</td> <td class="col-params">7B</td> <td class="col-mit">68.6</td> <td class="col-mit">47.5</td> <td class="col-crossner">63.1</td> <td class="col-crossner">68.2</td> <td class="col-crossner">75.7</td> <td class="col-crossner">69.4</td> <td class="col-crossner">69.9</td> <td class="col-buster" style="color: red;">23.6</td> <td class="col-avg">60.8</td> </tr> <tr> <td class="col-model"><b>SLIMER</b></td> <td class="col-backbone"><b>LLaMA-3.1-Instruct</b></td> <td class="col-params"><b>8B</b></td> <td class="col-mit"><b>56.4</b></td> <td class="col-mit"><b>44.8</b></td> <td class="col-crossner"><b>55.6</b></td> <td class="col-crossner"><b>63.3</b></td> <td class="col-crossner"><b>68.8</b></td> <td class="col-crossner"><b>69.2</b></td> <td class="col-crossner"><b>67.3</b></td> <td class="col-buster"><b>45.9</b></td> <td class="col-avg"><b>59.0</b></td> </tr> </tbody> </table> </body> </html> <!DOCTYPE html> <html lang="en"> <head> <meta charset="UTF-8"> <meta name="viewport" content="width=device-width, initial-scale=1.0"> <title>JSON Template</title> <style> body { font-family: Arial, sans-serif; line-height: 1.6; padding: 20px; } .description { font-weight: bold; color: #333; margin-bottom: 10px; } .template { background-color: #f0f0f0; padding: 10px; border-radius: 5px; margin-bottom: 20px; } .highlight-orange { color: orange; font-weight: bold; } </style> </head> <body> <div class="description">JSON SLIMER prompt</div> <div class="template"> <pre>{ "description": "SLIMER prompt", "prompt_input": "<|start_header_id|>system<|end_header_id|>\n\nYou are an expert in Named Entity Recognition designed to output JSON only.<|eot_id|>\n<|start_header_id|>user<|end_header_id|>\n\nYou are given a text chunk (delimited by triple quotes) and an instruction.\nRead the text and answer to the instruction in the end.\n\"\"\"\n{<span class="highlight-orange">input</span>}\n\"\"\"\nInstruction: Extract the Named Entities of type {<span class="highlight-orange">NE_name</span>} from the text chunk you have read. You are given a DEFINITION and some GUIDELINES.\nDEFINITION: {<span class="highlight-orange">definition</span>}\nGUIDELINES: {<span class="highlight-orange">guidelines</span>}\nReturn a JSON list of instances of this Named Entity type (for example [\"text_span_1\", \"text_span_2\"]. Return an empty list [] if no instances are present. Return only the JSON list, no further motivations or introduction to the answer.<|eot_id|>\n<|start_header_id|>assistant<|end_header_id|>\n\n" }</pre> </div> </body> </html>from vllm import LLM, SamplingParams
vllm_model = LLM(model="expertai/SLIMER-LLaMA3")
sampling_params = SamplingParams(temperature=0, max_tokens=128)
prompts = [prompter.generate_prompt(instruction, input) for instruction, input in instruction_input_pairs]
responses = vllm_model.generate(prompts, sampling_params)
Citation
If you find SLIMER useful in your research or work, please cite the following paper:
@misc{zamai2024lessinstructmoreenriching,
title={Show Less, Instruct More: Enriching Prompts with Definitions and Guidelines for Zero-Shot NER},
author={Andrew Zamai and Andrea Zugarini and Leonardo Rigutini and Marco Ernandes and Marco Maggini},
year={2024},
eprint={2407.01272},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2407.01272},
}
- Task
- Text Generation
- Type
- llama
- License
- llama3.1
- Languages
- en
- Featherless Ai
- live
- Created
- Oct 8, 2024
- Updated
- Nov 5, 2024