diff --git a/assets/multi_modal_files/nutrition_data.json b/assets/multi_modal_files/nutrition_data.json
index 5c57cbb..0a1d11c 100644
--- a/assets/multi_modal_files/nutrition_data.json
+++ b/assets/multi_modal_files/nutrition_data.json
@@ -28,5 +28,5 @@
"image_url": "https://photos-us.bazaarvoice.com/photo/2/cGhvdG86Ymx1ZWRpYW1vbmRncm93ZXJz/b20cd90a-4e8b-5cfb-aabb-88df46c19be7",
"title": "Chocolate Almond Milk nutrition facts",
"content": "Chocolate Almond Milk nutrition facts per cup"
- }
+ }
]
diff --git a/config/empty_image_ret_pipeline.yaml b/config/empty_image_ret_pipeline.yaml
new file mode 100644
index 0000000..9872fb4
--- /dev/null
+++ b/config/empty_image_ret_pipeline.yaml
@@ -0,0 +1,58 @@
+components:
+ prompt_builder:
+ init_parameters:
+ required_variables: null
+ template: "{% for document in documents %}\nImage: <|image_\nThis image is relevant to the question. \n{% endfor %}"
+ variables: null
+ type: fastrag.prompt_builders.multi_modal_prompt_builder.MultiModalPromptBuilder
+ retriever:
+ init_parameters:
+ document_store:
+ init_parameters:
+ bm25_algorithm: BM25L
+ bm25_parameters: {}
+ bm25_tokenization_regex: (?u)\b\w\w+\b
+ embedding_similarity_function: dot_product
+ index: fd8404e2-143c-4b33-9842-34aa23026912
+ type: haystack.document_stores.in_memory.document_store.InMemoryDocumentStore
+ filter_policy: replace
+ filters: null
+ return_embedding: false
+ scale_score: false
+ top_k: 1
+ type: haystack.components.retrievers.in_memory.embedding_retriever.InMemoryEmbeddingRetriever
+ text_embedder:
+ init_parameters:
+ batch_size: 32
+ config_kwargs: null
+ device:
+ device: cpu
+ type: single
+ embedding_separator: '
+
+ '
+ meta_fields_to_embed: []
+ model: openai/clip-vit-large-patch14
+ model_kwargs: null
+ normalize_embeddings: false
+ precision: float32
+ prefix: ''
+ progress_bar: true
+ suffix: ''
+ token:
+ env_vars:
+ - HF_API_TOKEN
+ - HF_TOKEN
+ strict: false
+ type: env_var
+ tokenizer_kwargs: null
+ truncate_dim: null
+ trust_remote_code: false
+ type: fastrag.embedders.image_embedders.SentenceTransformersImageTextEmbedder
+connections:
+- receiver: retriever.query_embedding
+ sender: text_embedder.embedding
+- receiver: prompt_builder.documents
+ sender: retriever.documents
+max_runs_per_component: 100
+metadata: {}
diff --git a/config/empty_ret_llama_32.yaml b/config/empty_ret_llama_32.yaml
new file mode 100644
index 0000000..efadcc8
--- /dev/null
+++ b/config/empty_ret_llama_32.yaml
@@ -0,0 +1,41 @@
+components:
+ embedder:
+ init_parameters:
+ batch_size: 32
+ device:
+ device: cpu
+ type: single
+ model: sentence-transformers/all-MiniLM-L6-v2
+ normalize_embeddings: false
+ prefix: ''
+ progress_bar: true
+ suffix: ''
+ token:
+ env_vars:
+ - HF_API_TOKEN
+ strict: false
+ type: env_var
+ type: haystack.components.embedders.sentence_transformers_text_embedder.SentenceTransformersTextEmbedder
+ prompt_builder:
+ init_parameters:
+ template: "```{% for document in documents %}\nImage: <|image|>\nThis image shows: {{ document.content }} \n{% endfor %}```\nThought: "
+ type: fastrag.prompt_builders.multi_modal_prompt_builder.MultiModalPromptBuilder
+ retriever:
+ init_parameters:
+ document_store:
+ init_parameters:
+ bm25_algorithm: BM25L
+ bm25_parameters: {}
+ bm25_tokenization_regex: (?u)\b\w\w+\b
+ embedding_similarity_function: dot_product
+ type: haystack.document_stores.in_memory.document_store.InMemoryDocumentStore
+ filters: null
+ return_embedding: false
+ scale_score: false
+ top_k: 1
+ type: haystack.components.retrievers.in_memory.embedding_retriever.InMemoryEmbeddingRetriever
+connections:
+- receiver: retriever.query_embedding
+ sender: embedder.embedding
+- receiver: prompt_builder.documents
+ sender: retriever.documents
diff --git a/config/vis_agent_phi_35.yaml b/config/vis_agent_phi_35.yaml
new file mode 100644
index 0000000..2381655
--- /dev/null
+++ b/config/vis_agent_phi_35.yaml
@@ -0,0 +1,24 @@
+chat_model:
+ generator_kwargs:
+ model: microsoft/Phi-3.5-vision-instruct
+ task: "image-to-text"
+ generation_kwargs:
+ max_new_tokens: 300
+ do_sample: false
+ huggingface_pipeline_kwargs:
+ torch_dtype: torch.bfloat16
+ trust_remote_code: true
+ _attn_implementation: eager
+ generator_class: fastrag.generators.llava.Phi35VisionHFGenerator
+tools:
+ - type: doc_with_image
+ params:
+ name: "docRetriever"
+ description: 'useful for when you need to retrieve images and text to answer questions. Use the following format: {{ "input": [your tool input here ] }}.'
+ pipeline_or_yaml_file: "config/empty_retrieval_pipeline.yaml"
+system_tools:
+ - type: doc_with_image_index_from_provider
+ params:
+ name: "docIndex"
+ pipeline_or_yaml_file: "config/empty_index_pipeline.yaml"
+ tool_provider_name: docRetriever
diff --git a/config/vis_agent_phi_35_image.yaml b/config/vis_agent_phi_35_image.yaml
new file mode 100644
index 0000000..8299085
--- /dev/null
+++ b/config/vis_agent_phi_35_image.yaml
@@ -0,0 +1,24 @@
+chat_model:
+ generator_kwargs:
+ model: microsoft/Phi-3.5-vision-instruct
+ task: "image-to-text"
+ generation_kwargs:
+ max_new_tokens: 300
+ do_sample: false
+ huggingface_pipeline_kwargs:
+ torch_dtype: torch.bfloat16
+ trust_remote_code: true
+ _attn_implementation: eager
+ generator_class: fastrag.generators.llava.Phi35VisionHFGenerator
+tools:
+ - type: doc_with_image
+ params:
+ name: imageRetriever
+ description: 'useful for when you need to retrieve images to answer questions. Use the following format: {{ "input": [your tool input here ] }}.'
+ pipeline_or_yaml_file: "config/empty_image_ret_pipeline.yaml"
+system_tools:
+ - type: image
+ params:
+ name: "docIndex"
+ model_name_or_path: "openai/clip-vit-large-patch14"
+ tool_provider_name: imageRetriever
diff --git a/config/visual_chat_agent.yaml b/config/visual_chat_agent.yaml
index 2381655..e765bb7 100644
--- a/config/visual_chat_agent.yaml
+++ b/config/visual_chat_agent.yaml
@@ -1,6 +1,6 @@
chat_model:
generator_kwargs:
- model: microsoft/Phi-3.5-vision-instruct
+ model: meta-llama/Llama-3.2-11B-Vision-Instruct
task: "image-to-text"
generation_kwargs:
max_new_tokens: 300
@@ -8,14 +8,13 @@ chat_model:
huggingface_pipeline_kwargs:
torch_dtype: torch.bfloat16
trust_remote_code: true
- _attn_implementation: eager
- generator_class: fastrag.generators.llava.Phi35VisionHFGenerator
+ generator_class: fastrag.generators.llava.LlamaHFGenerator
tools:
- type: doc_with_image
params:
name: "docRetriever"
description: 'useful for when you need to retrieve images and text to answer questions. Use the following format: {{ "input": [your tool input here ] }}.'
- pipeline_or_yaml_file: "config/empty_retrieval_pipeline.yaml"
+ pipeline_or_yaml_file: "config/empty_ret_llama_32.yaml"
system_tools:
- type: doc_with_image_index_from_provider
params:
diff --git a/fastrag/agents/base.py b/fastrag/agents/base.py
index 28e2ce8..81caba0 100644
--- a/fastrag/agents/base.py
+++ b/fastrag/agents/base.py
@@ -421,3 +421,7 @@ def create_agent_step(self, max_steps: Optional[int] = None) -> AgentStep:
return AgentStep(
max_steps=max_steps or self.max_steps, final_answer_pattern=self.final_answer_pattern
)
+
+ def clear(self):
+ self.memory.clear()
+ self.tm.clear_tool_history()
diff --git a/fastrag/agents/create_agent.py b/fastrag/agents/create_agent.py
index 67ce071..a8102a1 100644
--- a/fastrag/agents/create_agent.py
+++ b/fastrag/agents/create_agent.py
@@ -82,7 +82,7 @@ def get_generator(chat_model_config):
if not tokenizer.pad_token:
tokenizer.pad_token = tokenizer.eos_token
- stop_word_list = ["Observation:", "<|eot_id|>", "<|end|>"]
+ stop_word_list = ["Tool Output", "Observation:", "<|eot_id|>", "<|end|>"]
sw = StopWordsByTextCriteria(tokenizer=tokenizer, stop_words=stop_word_list, device="cpu")
if "generation_kwargs" not in chat_model_config["generator_kwargs"]:
diff --git a/fastrag/agents/tools/image_utils.py b/fastrag/agents/tools/image_utils.py
new file mode 100644
index 0000000..d2c7727
--- /dev/null
+++ b/fastrag/agents/tools/image_utils.py
@@ -0,0 +1,21 @@
+import base64
+from io import BytesIO
+
+import requests
+
+
+def get_base64_from_url(image_url):
+ if image_url.startswith("http"):
+ response = requests.get(
+ image_url,
+ headers={
+ "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/109.0.0.0 Safari/537.36",
+ },
+ )
+ buffered = BytesIO(response.content)
+ img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
+ else:
+ with open(image_url, "rb") as image_file:
+ img_str = base64.b64encode(image_file.read())
+
+ return img_str
diff --git a/fastrag/agents/tools/tools.py b/fastrag/agents/tools/tools.py
index a0ca2ba..3243a9d 100644
--- a/fastrag/agents/tools/tools.py
+++ b/fastrag/agents/tools/tools.py
@@ -6,7 +6,9 @@
from tqdm import tqdm
from fastrag.agents.base import Tool
+from fastrag.agents.tools.image_utils import get_base64_from_url
from fastrag.agents.utils import Color, load_text
+from fastrag.embedders.image_embedders import SentenceTransformersImageEmbedder
COMPONENT_WITH_STORE = "retriever"
@@ -190,10 +192,55 @@ def __init__(
def example_to_doc(self, ex):
return Document(
- content=ex["content"], meta={"title": ex["title"], "image_url": ex["image_url"]}
+ content=ex["content"],
+ meta={
+ "title": ex["title"],
+ "image_url": ex["image_url"],
+ "image_base64": get_base64_from_url(ex["image_url"]),
+ },
)
+class ImageHaystackIndexTool(Tool):
+ def __init__(
+ self,
+ name: str,
+ description: str = "",
+ logging_color: Color = Color.YELLOW,
+ model_name_or_path=None,
+ tool_provider_map: Dict[str, Tool] = None,
+ tool_provider_name: str = None,
+ ):
+ # get the store from the correct tool with the document store to use
+ document_store = tool_provider_map[tool_provider_name].get_store()
+ super().__init__(
+ name=name,
+ description=description,
+ logging_color=logging_color,
+ )
+ self.doc_embedder = SentenceTransformersImageEmbedder(model=model_name_or_path)
+ self.doc_embedder.warm_up()
+
+ self.document_store = document_store
+
+ def run(self, tool_input: Union[str, List[dict]], params: Optional[dict] = None) -> str:
+ if isinstance(tool_input, str):
+ tool_input = json.loads(tool_input)
+
+ elif isinstance(tool_input, dict) and "docs" in tool_input:
+ tool_input = tool_input["docs"]
+
+ docs = [
+ Document(
+ content=element["content"],
+ meta=dict(image_base64=get_base64_from_url(element["content"])),
+ )
+ for element in tool_input
+ ]
+ docs_with_embeddings = self.doc_embedder.run(docs)
+ self.document_store.write_documents(docs_with_embeddings["documents"])
+
+
class DocWithImageFromProvidersHaystackIndexTool(DocWithImageHaystackIndexTool):
def __init__(
self,
@@ -220,4 +267,5 @@ def __init__(
"doc_with_image_index": DocWithImageHaystackIndexTool,
"doc_with_image_index_from_provider": DocWithImageFromProvidersHaystackIndexTool,
"doc": HaystackQueryTool,
+ "image": ImageHaystackIndexTool,
}
diff --git a/fastrag/embedders/image_embedders.py b/fastrag/embedders/image_embedders.py
new file mode 100644
index 0000000..eb67a81
--- /dev/null
+++ b/fastrag/embedders/image_embedders.py
@@ -0,0 +1,68 @@
+import base64
+from io import BytesIO
+from typing import List
+
+import torch
+from haystack import Document, component
+from haystack.components.embedders import (
+ SentenceTransformersDocumentEmbedder,
+ SentenceTransformersTextEmbedder,
+)
+from PIL import Image
+from transformers import CLIPModel, CLIPProcessor
+
+
+def base64_to_image(base_64_input):
+ bytes_io = BytesIO(base64.b64decode(base_64_input))
+ return Image.open(bytes_io)
+
+
+class BaseSentenceTransformersImageEmbedder(SentenceTransformersDocumentEmbedder):
+ def warm_up(self):
+ """
+ Initializes the component.
+ """
+
+ # Load CLIP model
+ self.embeding_backend = CLIPModel.from_pretrained(self.model)
+ self.processor = CLIPProcessor.from_pretrained(self.model)
+
+
+@component
+class SentenceTransformersImageEmbedder(
+ BaseSentenceTransformersImageEmbedder, SentenceTransformersDocumentEmbedder
+):
+ @component.output_types(documents=List[Document])
+ def run(self, documents: List[Document]):
+ images = [base64_to_image(doc.meta["image_base64"]) for doc in documents]
+ inputs = self.processor(images=images, return_tensors="pt", padding=True)
+ with torch.no_grad():
+ embeddings = self.embeding_backend.get_image_features(**inputs)
+
+ for doc, emb in zip(documents, embeddings):
+ doc.embedding = emb
+
+ return {"documents": documents}
+
+
+@component
+class SentenceTransformersImageTextEmbedder(
+ BaseSentenceTransformersImageEmbedder, SentenceTransformersTextEmbedder
+):
+ @component.output_types(embedding=List[float])
+ def run(self, text: str):
+ """
+ Embed a single string.
+
+ :param text:
+ Text to embed.
+
+ :returns:
+ A dictionary with the following keys:
+ - `embedding`: The embedding of the input text.
+ """
+ inputs = self.processor(text=text, return_tensors="pt", padding=True)
+ with torch.no_grad():
+ text_outputs = self.embeding_backend.get_text_features(**inputs)
+
+ return {"embedding": text_outputs.tolist()[0]}
diff --git a/fastrag/generators/image_caption_generator.py b/fastrag/generators/image_caption_generator.py
new file mode 100644
index 0000000..cbafc41
--- /dev/null
+++ b/fastrag/generators/image_caption_generator.py
@@ -0,0 +1,73 @@
+import torch
+from PIL import Image
+from transformers import AutoModelForCausalLM, AutoModelForVision2Seq, AutoProcessor
+from transformers.image_utils import load_image
+
+
+def get_default_device():
+ if torch.cuda.is_available():
+ return "cuda"
+ if torch.xpu.is_available():
+ return "xpu"
+ return "cpu"
+
+
+class ImageCaptionGenerator:
+ def __init__(self, model_name_or_path="microsoft/caption-large"):
+ print("Loading Image Caption model ...")
+ self.model_name_or_path = model_name_or_path
+ self.device = get_default_device()
+ self.processor = AutoProcessor.from_pretrained(model_name_or_path)
+
+ def caption(self, path):
+ raise NotImplementedError()
+
+
+class GitImageCaptionGenerator(ImageCaptionGenerator):
+ def __init__(self, model_name_or_path="microsoft/git-large-r-textcaps"):
+ super().__init__(model_name_or_path)
+ self.model = AutoModelForCausalLM.from_pretrained(model_name_or_path).to(self.device)
+
+ def caption(self, path):
+ image = Image.open(path)
+
+ pixel_values = self.processor(images=image, return_tensors="pt").pixel_values
+ generated_ids = self.model.generate(
+ pixel_values=pixel_values.to(self.model.device), max_length=30, do_sample=False
+ )
+ generated_caption = self.processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
+ return generated_caption
+
+
+class VLLMImageCaptionGenerator(ImageCaptionGenerator):
+ def __init__(self, model_name_or_path="HuggingFaceTB/SmolVLM-Instruct"):
+ super().__init__(model_name_or_path)
+
+ self.model = AutoModelForVision2Seq.from_pretrained(
+ model_name_or_path,
+ torch_dtype=torch.bfloat16,
+ _attn_implementation="flash_attention_2" if self.device == "cuda" else "eager",
+ )
+
+ def caption(self, path):
+ # Load images
+ image1 = load_image(path)
+
+ # Create input messages
+ messages = [
+ {
+ "role": "user",
+ "content": [
+ {"type": "image"},
+ {"type": "text", "text": "Can you describe the image?"},
+ ],
+ },
+ ]
+
+ # Prepare inputs
+ prompt = self.processor.apply_chat_template(messages, add_generation_prompt=True)
+ inputs = self.processor(text=prompt, images=[image1], return_tensors="pt")
+ inputs = inputs.to(self.model.device)
+ # Generate outputs
+ generated_ids = self.model.generate(**inputs, max_new_tokens=20, do_sample=False)
+ return self.processor.batch_decode(generated_ids[:, inputs["input_ids"].shape[1] :])[0]
diff --git a/fastrag/generators/llava.py b/fastrag/generators/llava.py
index 6070114..476b860 100644
--- a/fastrag/generators/llava.py
+++ b/fastrag/generators/llava.py
@@ -113,7 +113,13 @@ def __init__(
del self.generation_kwargs["stopping_criteria"]
self.processor = AutoProcessor.from_pretrained(model)
- self.image_token = "<image>"
+ image_tokens = [
+ v
+ for v in self.processor.tokenizer.added_tokens_decoder.values()
+ if "image" in v.content
+ ]
+ assert len(image_tokens) > 0, "No image token found in the tokenizer"
+ self.image_token = image_tokens[0].content
@component.output_types(replies=List[str])
def run(
@@ -208,6 +214,92 @@ def get_user_text(self, chat_snippet):
return user_text
+class LlamaHFGenerator(LlavaHFGenerator):
+ @component.output_types(replies=List[str])
+ def run(
+ self,
+ prompt: str,
+ images: List[str] = None,
+ generation_kwargs: Optional[Dict[str, Any]] = None,
+ ):
+ """
+ Run the text generation model on the given prompt.
+
+ :param prompt:
+ A string representing the prompt.
+ :param images:
+ A list of base64 strings representing the input images.
+ :param generation_kwargs:
+ Additional keyword arguments for text generation.
+
+ :returns:
+ A dictionary containing the generated replies.
+ - replies: A list of strings representing the generated replies.
+ - raw_images: A list of the raw PIL images.
+ """
+ if self.pipeline is None:
+ raise RuntimeError(
+ "The generation model has not been loaded. Please call warm_up() before running."
+ )
+
+ if not prompt:
+ return {"replies": []}
+
+ # merge generation kwargs from init method with those from run method
+ updated_generation_kwargs = {**self.generation_kwargs, **(generation_kwargs or {})}
+
+ raw_images = None
+ if images:
+ raw_images = [base64_to_image(img) for img in images]
+
+ present_image_token_count = prompt.count(self.image_token)
+ image_token_count_diff = len(images) - present_image_token_count
+
+ # check if we need to add additional image tokens
+ if image_token_count_diff > 0:
+ image_token_full_str = " ".join(
+ [self.image_token for _ in range(image_token_count_diff)]
+ )
+ prompt = f"Current Images: {image_token_full_str}\n" + prompt
+
+ print(f"USING {len(raw_images)=}!!!!!!!!")
+ inputs = self.processor(
+ images=raw_images, text=prompt, add_special_tokens=False, return_tensors="pt"
+ )
+ else:
+ inputs = self.processor(
+ images=None, text=prompt, add_special_tokens=False, return_tensors="pt"
+ )
+
+ updated_generation_kwargs["max_length"] = updated_generation_kwargs.get("max_length", 32000)
+ print(f"{prompt=}")
+ stop_strings = self.stopping_criteria_list[0].stop_words_text
+ output = self.pipeline.model.generate(
+ **inputs,
+ stop_strings=stop_strings,
+ tokenizer=self.processor.tokenizer,
+ **updated_generation_kwargs,
+ )
+
+ replies = self.processor.batch_decode(
+ output[:, inputs["input_ids"].shape[1] :], skip_special_tokens=True
+ )
+
+ gen_stop_words = self.get_stop_words_from_kwargs()
+
+ stop_words = None
+ stop_words = gen_stop_words if gen_stop_words else self.stop_words
+ if stop_words:
+ # the output of the pipeline includes the stop word
+ replies = [
+ reply.replace(stop_word, "").rstrip()
+ for reply in replies
+ for stop_word in stop_words
+ ]
+
+ return {"replies": replies, "raw_images": raw_images}
+
+
class Phi35VisionHFGenerator(HuggingFaceLocalGenerator):
"""
Generator based on a Phi35 Hugging Face model loaded.
diff --git a/fastrag/prompt_builders/multi_modal_prompt_builder.py b/fastrag/prompt_builders/multi_modal_prompt_builder.py
index 3f17528..6ccada2 100644
--- a/fastrag/prompt_builders/multi_modal_prompt_builder.py
+++ b/fastrag/prompt_builders/multi_modal_prompt_builder.py
@@ -31,22 +31,10 @@ def run(self, **kwargs):
- `prompt`: The updated prompt text after rendering the prompt template.
"""
prompt_dict = {"prompt": self.template.render(kwargs)}
- prompt_dict["images"] = [
- self.get_base64_from_url(doc.meta["image_url"]) for doc in kwargs["documents"]
- ]
- return prompt_dict
+ prompt_dict["images"] = [doc.meta["image_base64"] for doc in kwargs["documents"]]
- def get_base64_from_url(self, image_url):
- response = requests.get(
- image_url,
- headers={
- "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/109.0.0.0 Safari/537.36",
- },
- )
- buffered = BytesIO(response.content)
- img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
- return img_str
+ return prompt_dict
component._component(MultiModalPromptBuilder)
diff --git a/fastrag/ui/chainlit_multi_modal_agent.py b/fastrag/ui/chainlit_multi_modal_agent.py
index 9c1942e..4390781 100644
--- a/fastrag/ui/chainlit_multi_modal_agent.py
+++ b/fastrag/ui/chainlit_multi_modal_agent.py
@@ -45,12 +45,12 @@
@cl.on_chat_end
def chat_end():
# clear memory
- agent.memory.clear()
+ agent.clear()
def add_images_to_message(additional_params):
image_elements = []
- last_image = [additional_params["images"][-1]]
+ last_image = additional_params["images"]
for image_base64_index, image_base64 in enumerate(last_image):
image_uuid = str(uuid.uuid4())
bytes_io = BytesIO(base64.b64decode(image_base64))
@@ -75,6 +75,16 @@ def parse_element(element, params):
if "json" in element.mime:
params["docs"] = json.load(open(element.path, "r"))
+ if any([image_suffix in element.mime for image_suffix in ["png", "jpeg", "jpg"]]):
+ if "docs" not in params:
+ params["docs"] = []
+
+ image_element = {
+ "content": element.path,
+ }
+
+ params["docs"].append(image_element)
+
# params for the agent
params = {}
@@ -95,8 +105,12 @@ def parse_element(element, params):
answer = agent_result["answers"][0].answer
# display retrieved image, if exists
- additional_params = agent.memory.get_additional_params()
- if "images" in additional_params and len(additional_params["images"]) > 0:
+ additional_params = agent.memory.list[-1].get("additional_params", None)
+ if (
+ additional_params
+ and "images" in additional_params
+ and len(additional_params["images"]) > 0
+ ):
image_elements = add_images_to_message(additional_params)
_ = await cl.Message(
diff --git a/setup.cfg b/setup.cfg
index c0790e1..a852d67 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -19,6 +19,7 @@ install_requires =
chainlit>=1.0.506
sentence-transformers>=2.3.0
events
+ pydantic==2.10.1
[options.extras_require]
dev =
diff --git a/setup.py b/setup.py
index 9636edc..30f8e29 100644
--- a/setup.py
+++ b/setup.py
@@ -30,7 +30,7 @@ def get_version(rel_path):
long_description_content_type="text/markdown",
url="https://github.com/IntelLabs/fastRAG",
license="Apache-2.0",
- python_requires=">=3.8, <3.12",
+ python_requires=">=3.8",
classifiers=[
"Programming Language :: Python :: 3",
"License :: OSI Approved :: Apache Software License",