Outputs

import json
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.llms.ollama import Ollama
from llama_index.llms.openai import OpenAI
from pydantic import BaseModel, Field, field_validator
from sympy import N

class Fact(BaseModel):
    """A fact from the text."""
    fact: str = Field(description="A fact from the text.")
    dates: list[str] | None = Field(description="The dates the fact is relevant to.")
    surrounding_text: str | None = Field(description="The text surrounding the fact.")

class Facts(BaseModel):
    """A list of facts from the text."""
    facts: list[Fact] = Field(description="A list of facts.")

    @field_validator("facts", mode="before")
    def validate_facts(cls, v):
        if not v:
            raise ValueError("Facts list is empty")
        if isinstance(v, str):
            v = json.loads(v)
            if isinstance(v, list):
                return [Fact(**fact) for fact in v]

        return v




llm = Ollama(model="llama3.1:latest", request_timeout=120.0)
# llm = Ollama(model="mistral-nemo:latest", request_timeout=120.0)
# llm = OpenAI(model="gpt-4o-mini")

documents = SimpleDirectoryReader("./docs/docs/examples/data/paul_graham").load_data()

index = VectorStoreIndex.from_documents(documents)

query_engine = index.as_query_engine(llm=llm, output_cls=Facts)

response = query_engine.query("What did the author do growing up? Extract the facts.")

breakpoint()
print(type(response))
print(response)
# this will fail if the llm failed to write a valid object
print(response.facts)

llm-1        | Terminating the parser. Please open an issue at 
llm-1        | https://github.com/noamgat/lm-format-enforcer/issues with the prefix and CharacterLevelParser parameters
llm-1        | Traceback (most recent call last):
llm-1        |   File "/opt/nim/llm/.venv/lib/python3.10/site-packages/lmformatenforcer/tokenenforcer.py", line 96, in _compute_allowed_tokens
llm-1        |     self._collect_allowed_tokens(state.parser, self.tokenizer_tree.root, allowed_tokens, shortcut_key)
llm-1        |   File "/opt/nim/llm/.venv/lib/python3.10/site-packages/lmformatenforcer/tokenenforcer.py", line 144, in _collect_allowed_tokens
llm-1        |     self._collect_allowed_tokens(next_parser, next_tree_node, allowed_tokens, None)
llm-1        |   File "/opt/nim/llm/.venv/lib/python3.10/site-packages/lmformatenforcer/tokenenforcer.py", line 142, in _collect_allowed_tokens
llm-1        |     next_parser = parser.add_character(character)
llm-1        |   File "/opt/nim/llm/.venv/lib/python3.10/site-packages/lmformatenforcer/characterlevelparser.py", line 152, in add_character
llm-1        |     updated_parser = parser.add_character(new_character)
llm-1        |   File "/opt/nim/llm/.venv/lib/python3.10/site-packages/lmformatenforcer/jsonschemaparser.py", line 74, in add_character
llm-1        |     updated_parser.object_stack[receiving_idx] = updated_parser.object_stack[receiving_idx].add_character(new_character)
llm-1        |   File "/opt/nim/llm/.venv/lib/python3.10/site-packages/lmformatenforcer/jsonschemaparser.py", line 347, in add_character
llm-1        |     self.current_key_parser = get_parser(
llm-1        |   File "/opt/nim/llm/.venv/lib/python3.10/site-packages/lmformatenforcer/jsonschemaparser.py", line 187, in get_parser
llm-1        |     return get_parser(parsing_state, merged_schema)
llm-1        |   File "/opt/nim/llm/.venv/lib/python3.10/site-packages/lmformatenforcer/jsonschemaparser.py", line 221, in get_parser
llm-1        |     raise ValueError("No definitions found in schema")
llm-1        | ValueError: No definitions found in schema
llm-1        | WARNING 2024-11-15 21:13:55.219 serving_chat.py:263] A tool call was detected, but the resulting JSON was unparseable. Try increasing `max_tokens`: Unterminated string starting at: line 1 column 94 (char 93)

llama-index==0.11.23
llama-index-embeddings-nvidia==0.2.5
llama-index-embeddings-ollama==0.3.1
llama-index-vector-stores-chroma==0.3.0
llama-index-llms-nvidia==0.2.6
llama-index-llms-ollama==0.3.0

data-process-1   |   File "/usr/local/lib/python3.11/site-packages/llama_index/llms/nvidia/base.py", line 290, in get_tool_calls_from_response
data-process-1   |     raise ValueError(
data-process-1   | ValueError: Expected at least one tool call, but got 0 tool calls.

from llama_index.core.program.function_calling import get_function_tool

retriever = index.as_retriever(similarity_top_k=2, ...)

nodes = retriever.retrieve("query")
context_str = "\n\n".join([x.text for x in nodes])

tool = get_function_tool(OutputCLS)
# could also use chat_history=chat_history instead, but here we assume a single message
resp = llm.chat_with_tools(
  [tool], 
  user_msg=f"Given some context and a user query, do a thing.\nContext: {context_str}\n\nQuery: {query}"
)

tool_calls = llm.get_tool_calls_from_response(resp, error_on_no_tool_call=False)
output_objs
for tool_call in tool_calls:
  if tool_call.tool_name == tool.metadata.name:
    output_objs.append(OutputCLS(**tool_call.tool_kwargs)

print(outout_objs