start to integrate

external_components/llm_extraction/Dockerfile

@@ -0,0 +1,24 @@
+FROM condaforge/mambaforge:24.3.0-0
+SHELL ["/bin/bash", "--login", "-c"]
+# Set working directory
+WORKDIR /app
+
+COPY external_components/rtransparent/environment.yaml /app
+
+# Create the environment
+RUN conda env create -f environment.yaml
+
+# Ensure the conda environment is activated
+RUN echo "source /opt/conda/etc/profile.d/conda.sh && conda activate osm" | tee -a ~/.bashrc /etc/profile /etc/profile.d/conda.sh /etc/skel/.bashrc /etc/skel/.profile > /dev/null
+
+RUN R -e '\
+devtools::install_github("quest-bih/oddpub",ref="c5b091c7e82ed6177192dc380a515b3dc6304863"); \
+devtools::install_github("serghiou/rtransparent", build_vignettes = F)'
+
+# # Copy the project files and install the package
+COPY external_components/rtransparent/app.py /app
+
+# Make entrypoint etc. convenient for users
+COPY external_components/_entrypoint.sh /usr/local/bin/_entrypoint.sh
+ENTRYPOINT ["/usr/local/bin/_entrypoint.sh"]
+CMD ["fastapi", "dev", "--host", "0.0.0.0", "--port", "8071"]

external_components/llm_extraction/app.py

@@ -0,0 +1,77 @@
+import logging
+
+from fastapi import FastAPI, File, HTTPException, Query, UploadFile
+from llama_index.core import ChatPromptTemplate
+from llama_index.core.llms import LLM, ChatMessage
+from llama_index.llms.openai import OpenAI
+from llama_index.program.openai import OpenAIPydanticProgram
+
+# from pydantic import BaseModel, Field
+from osm.schemas.metrics_schemas import LLMExtractor
+
+LLM_MODELS = {"gpt-4o-2024-08-06": OpenAI(model="gpt-4o-2024-08-06")}
+
+
+logger = logging.getLogger(__name__)
+app = FastAPI()
+
+
+def get_program(llm: LLM) -> OpenAIPydanticProgram:
+    prompt = ChatPromptTemplate(
+        message_templates=[
+            ChatMessage(
+                role="system",
+                content=(
+                    "You are an expert at extracting information from scientific publications with a keen eye for details that when combined together allows you to summarize aspects of the publication"
+                ),
+            ),
+            ChatMessage(
+                role="user",
+                content=(
+                    "The llm model is {llm_model}. The publication in xml follows below:\n"
+                    "------\n"
+                    "{xml_content}\n"
+                    "------"
+                ),
+            ),
+        ]
+    )
+
+    program = OpenAIPydanticProgram.from_defaults(
+        output_cls=LLMExtractor,
+        llm=llm,
+        prompt=prompt,
+        verbose=True,
+    )
+    return program
+
+
+def extract_with_llm(xml_content: bytes, llm: LLM) -> LLMExtractor:
+    program = get_program(llm=llm)
+    return program(xml_content=xml_content, llm_model=llm.model)
+
+
+def llm_metric_extraction(
+    xml_content: bytes,
+    llm_model: str,
+):
+    return extract_with_llm(xml_content, LLM_MODELS[llm_model])
+
+
+@app.post("/extract-metrics/", response_model=LLMExtractor)
+async def extract_metrics(
+    file: UploadFile = File(...), llm_model: str = Query("other")
+):
+    try:
+        xml_content = await file.read()
+        if not xml_content:
+            raise NotImplementedError(
+                """For now the XML content must be provided. Check the output of
+                the parsing stage."""
+            )
+        metrics = llm_metric_extraction(xml_content, llm_model)
+        logger.info(metrics)
+        return metrics
+
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))

external_components/llm_extraction/environment.yaml

@@ -0,0 +1,47 @@
+name: osm
+channels:
+  - conda-forge
+  - nodefaults
+dependencies:
+  - fastapi
+  - lxml
+  - pandas
+  - pip
+  - psutil
+  - python
+  - requests
+  - rpy2
+  - uvicorn
+  # Dependencies for rtransparent
+  - r-crul
+  - r-devtools
+  - r-dplyr
+  - r-furrr
+  - r-future
+  - r-globals
+  - r-hoardr
+  - r-httpcode
+  - r-lazyeval
+  - r-lubridate
+  - r-magrittr
+  - r-pbapply
+  - r-pdftools
+  - r-plyr
+  - r-purrr
+  - r-qpdf
+  - r-readr
+  # - r-rentrez
+  - r-rlang
+  - r-stringr
+  - r-tibble
+  - r-tidyr
+  - r-tidyselect
+  - r-timechange
+  - r-tokenizers
+  - r-triebeard
+  - r-urltools
+  - r-utf8
+  # - r-XML
+  - r-xml2
+  - pip:
+    - metapub

osm/pipeline/extractors.py

@@ -38,6 +38,28 @@ def _run(self, data: bytes, parser: str = None) -> dict:
             response.raise_for_status()
 
 
+class LLMExtractor(Component):
+    def _run(self, data: bytes, llm_model: str = None) -> dict:
+        self.sample = LongBytes(data)
+
+        # Prepare the file to be sent as a part of form data
+        files = {"file": ("input.xml", io.BytesIO(data), "application/xml")}
+
+        # Send the request with the file
+        response = requests.post(
+            "http://localhost:8072/extract-metrics/",
+            files=files,
+            params={"llm_model": llm_model},
+        )
+
+        if response.status_code == 200:
+            metrics = response.json()
+            return metrics
+        else:
+            logger.error(f"Error: {response.text}")
+            response.raise_for_status()
+
+
 # import psutil
 # # Adjust the logging level for rpy2
 # rpy2_logger = logging.getLogger("rpy2")