Lumixion-e1-70k-fncall-qlora

Lumixion-e1-70k-fncall-qlora

Lumixion is the first ever vast array of multi-modal function calling models easily available for usage. This is the first iteration finetuned on 70+ samples with qlora and many other optimizations. If you would like to work on real-world multi-modal AI join our discord: LINK

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
from transformers.generation import GenerationConfig

tokenizer = AutoTokenizer.from_pretrained("AgoraX/Lumixion-e1-70k-fncall-qlora",trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    "AgoraX/Lumixion-e1-70k-fncall-qlora", # path to the output directory
    device_map="cuda",
    trust_remote_code=True
).eval()



# 1st dialogue turn
query = tokenizer.from_list_format([
    {'image': 'https://images.rawpixel.com/image_800/cHJpdmF0ZS9sci9pbWFnZXMvd2Vic2l0ZS8yMDIzLTA4L3Jhd3BpeGVsX29mZmljZV8xNV9waG90b19vZl9hX2RvZ19ydW5uaW5nX3dpdGhfb3duZXJfYXRfcGFya19lcF9mM2I3MDQyZC0zNWJlLTRlMTQtOGZhNy1kY2Q2OWQ1YzQzZjlfMi5qcGc.jpg'}, # Either a local path or an url
    {'text': "What are the objects in the image? What animals are present? Are there any people in the image?"},
])
print("sending model to chat")
response, history = model.chat(tokenizer, query=query, history=None)
print(response)

## How to Get Started with the Model

output

[FUNCTION CALL]
{{
  'type': 'object',
  'properties': {{
    'objects': {{
      'type': 'array',
      'description': 'The objects present in the image.',
      'items': {{
        'type': 'string',
        'enum': ['dog', 'person', 'tree', 'path', 'sun']
      }}
    }},
    'animals': {{
      'type': 'array',
      'description': 'The animals present in the image.',
      'items': {{
        'type': 'string',
        'enum': ['dog']
      }}
    }},
    'people': {{
      'type': 'boolean',
      'description': 'Whether there are people in the image.',
      'enum': [true]
    }}
  }}
}}

[EXPECTED OUTPUT]
{{
  'objects': ['dog', 'person', 'tree', 'path', 'sun'],
  'animals': ['dog'],
  'people': true
}}

Model Details

Model Description

<!-- Provide a longer summary of what this model is. -->

• Developed by: Agora Research
• Model type: Vision Language Model
• Language(s) (NLP): English/Chinese
• Finetuned from model: Qwen-VL-Chat

Model Sources [optional]

<!-- Provide the basic links for the model. -->

• Repository: https://github.com/QwenLM/Qwen-VL
• Paper: https://arxiv.org/pdf/2308.12966.pdf

Uses

from transformers import AutoTokenizer, AutoModelForCausalLM
from transformers.generation import GenerationConfig

Note: The default behavior now has injection attack prevention off.

tokenizer = AutoTokenizer.from_pretrained("qwen/Qwen-VL-Chat",trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    "MODEL_PATH_HERE", # path to the output directory
    device_map="cuda",
    trust_remote_code=True
).eval()

Specify hyperparameters for generation (generation_config if transformers < 4.32.0)

#model.generation_config = GenerationConfig.from_pretrained("Qwen/Qwen-VL-Chat", trust_remote_code=True)


# 1st dialogue turn
query = tokenizer.from_list_format([
    {'image': 'https://images.rawpixel.com/image_800/cHJpdmF0ZS9sci9pbWFnZXMvd2Vic2l0ZS8yMDIzLTA4L3Jhd3BpeGVsX29mZmljZV8xNV9waG90b19vZl9hX2RvZ19ydW5uaW5nX3dpdGhfb3duZXJfYXRfcGFya19lcF9mM2I3MDQyZC0zNWJlLTRlMTQtOGZhNy1kY2Q2OWQ1YzQzZjlfMi5qcGc.jpg'}, # Either a local path or an url
    {'text': "What are the objects in the image? What animals are present? Are there any people in the image?"},
])
print("sending model to chat")
response, history = model.chat(tokenizer, query=query, history=None)
print(response)

Print Results

[FUNCTION CALL]
{{
  'type': 'object',
  'properties': {{
    'objects': {{
      'type': 'array',
      'description': 'The objects present in the image.',
      'items': {{
        'type': 'string',
        'enum': ['dog', 'person', 'tree', 'path', 'sun']
      }}
    }},
    'animals': {{
      'type': 'array',
      'description': 'The animals present in the image.',
      'items': {{
        'type': 'string',
        'enum': ['dog']
      }}
    }},
    'people': {{
      'type': 'boolean',
      'description': 'Whether there are people in the image.',
      'enum': [true]
    }}
  }}
}}

[EXPECTED OUTPUT]
{{
  'objects': ['dog', 'person', 'tree', 'path', 'sun'],
  'animals': ['dog'],
  'people': true
}}

Direct Use

Just send an image and ask a question in the text.

Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

(recommended) transformers >= 4.32.0

How to Get Started with the Model

query = tokenizer.from_list_format([
    {'image': 'https://images.rawpixel.com/image_800/cHJpdmF0ZS9sci9pbWFnZXMvd2Vic2l0ZS8yMDIzLTA4L3Jhd3BpeGVsX29mZmljZV8xNV9waG90b19vZl9hX2RvZ19ydW5uaW5nX3dpdGhfb3duZXJfYXRfcGFya19lcF9mM2I3MDQyZC0zNWJlLTRlMTQtOGZhNy1kY2Q2OWQ1YzQzZjlfMi5qcGc.jpg'}, # Either a local path or an url
    {'text': "QUESTIONS/QUERIES GO HERE"},
])

Training Details

Training Data

Custom Function Calling Dataset with 70k examples

Training Procedure

qlora for 3 epochs