Traceback (most recent call last): File "/home/robebs/gitlab/nlp/llama_indexx.py", line 43, in <module> response = sllm.complete(text) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index_instrumentation/dispatcher.py", line 335, in wrapper result = func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/core/llms/callbacks.py", line 435, in wrapped_llm_predict f_return_val = f(_self, *args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/core/llms/structured_llm.py", line 95, in complete return complete_fn(prompt, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/core/base/llms/generic_utils.py", line 184, in wrapper chat_response = func(messages, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index_instrumentation/dispatcher.py", line 335, in wrapper result = func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/core/llms/callbacks.py", line 175, in wrapped_llm_chat f_return_val = f(_self, messages, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/core/llms/structured_llm.py", line 63, in chat output = self.llm.structured_predict( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index_instrumentation/dispatcher.py", line 335, in wrapper result = func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/llms/openai/base.py", line 1119, in structured_predict return super().structured_predict( ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index_instrumentation/dispatcher.py", line 335, in wrapper result = func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/core/llms/llm.py", line 360, in structured_predict result = program(llm_kwargs=llm_kwargs, **prompt_args) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index_instrumentation/dispatcher.py", line 335, in wrapper result = func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/core/program/llm_program.py", line 98, in call response = self._llm.complete(formatted_prompt, **llm_kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index_instrumentation/dispatcher.py", line 335, in wrapper result = func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/llms/openai_like/base.py", line 168, in complete return super().complete(prompt, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index_instrumentation/dispatcher.py", line 335, in wrapper result = func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/core/llms/callbacks.py", line 435, in wrapped_llm_predict f_return_val = f(_self, *args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/llms/openai/base.py", line 428, in complete return complete_fn(prompt, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/llms/openai/base.py", line 114, in wrapper return retry(f)(self, *args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/tenacity/init.py", line 331, in wrapped_f return copy(f, *args, **kw) ^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/tenacity/init.py", line 470, in call do = self.iter(retry_state=retry_state) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/tenacity/init.py", line 371, in iter result = action(retry_state) ^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/tenacity/init.py", line 393, in <lambda> self._add_action_func(lambda rs: rs.outcome.result()) ^^^^^^^^^^^^^^^^^^^ File "/usr/lib/python3.12/concurrent/futures/_base.py", line 449, in result return self.__get_result() ^^^^^^^^^^^^^^^^^^^ File "/usr/lib/python3.12/concurrent/futures/_base.py", line 401, in __get_result raise self._exception File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/tenacity/init.py", line 473, in call result = fn(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/llama_index/llms/openai/base.py", line 605, in _complete response = client.completions.create( ^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/robebs/gitlab/nlp/.venv/lib/python3.12/site-packages/openai/_utils/_utils.py", line 286, in wrapper return func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ TypeError: Completions.create() got an unexpected keyword argument 'tool_choice'

Quick Fix

OpenAILike.as_structured_llm() assumes the underlying model supports function‑calling.
Because you created the client with is_function_calling_model=False, the wrapper still adds the tool_choice argument (used only for chat‑completion with function calling).
The OpenAI‑compatible server you are hitting does not accept that argument, so Completions.create() raises:

TypeError: Completions.create() got an unexpected keyword argument 'tool_choice'

Two ways to solve it

Both approaches stop the illegal tool_choice kw‑arg from being sent.

Example Code – Option A (recommended if the model supports it)

import os, json, httpx
from llama_index.llms.openai_like import OpenAILike
from llama_index.core.llms import StructuredLLM  # for typing only

# ------------------------------------------------------------------
# 1️⃣  Create the client **with function‑calling enabled**
# ------------------------------------------------------------------
client = OpenAILike(
    base_url=os.getenv("BASE_URL"),
    api_key=os.getenv("API_KEY"),
    http_client=httpx.Client(verify=False),
    model="gpt-oss-20b",
    # is_function_calling_model defaults to True – omit or set True
    is_function_calling_model=True,
)

# ------------------------------------------------------------------
# 2️⃣  Convert to a StructuredLLM (your Pydantic/Dataclass schema)
# ------------------------------------------------------------------
sllm: StructuredLLM = client.as_structured_llm(Invoice)   # replace Invoice with your schema

# ------------------------------------------------------------------
# 3️⃣  Run the request
# ------------------------------------------------------------------
text = "Please extract the invoice fields from the following text ..."
response = sllm.complete(text)

# ------------------------------------------------------------------
# 4️⃣  Inspect the JSON result
# ------------------------------------------------------------------
json_response = json.loads(response.text)
print(json.dumps(json_response, indent=2))

Example Code – Option B (if the model cannot do function calls)

import os, json, httpx
from llama_index.llms.openai_like import OpenAILike

client = OpenAILike(
    base_url=os.getenv("BASE_URL"),
    api_key=os.getenv("API_KEY"),
    http_client=httpx.Client(verify=False),
    model="gpt-oss-20b",
    is_function_calling_model=False,          # keep as False
    tool_choice="none",                       # <-- suppress the unsupported kwarg
)

sllm = client.as_structured_llm(Invoice)
response = sllm.complete(text)
print(json.dumps(json.loads(response.text), indent=2))

Why tool_choice="none" works – the StructuredLLM wrapper adds tool_choice only when it is non‑None. Setting it to "none" tells the wrapper to omit the argument, matching the server’s completion API.

Verification Steps

1. Run the script after applying one of the two fixes.
2. The traceback should disappear; you should receive a normal Response object.
3. response.text (or response.json()) should contain the structured output defined by your Invoice schema.

If you still see the same TypeError, double‑check that the client instance you are using is the one you modified (no stale imports) and that the library versions are at least:

llama-index-core >= 0.14.15
llama-index-llms-openai-like >= 0.6.0

Extra Tips

• Upgrade – newer llama-index releases (≥ 0.7.0 of llama-index-llms-openai-like) automatically suppress tool_choice when is_function_calling_model=False. Upgrading may remove the need for the manual tool_choice="none" flag.
• Environment – ensure the server you point BASE_URL to actually implements the OpenAI chat endpoint if you intend to use function calling; otherwise stick with the plain completions endpoint.
• Testing – write a tiny unit test that calls client.as_structured_llm and asserts that the request payload does not contain a tool_choice key when is_function_calling_model=False.
• Future‑proofing – when switching to a model that does support function calls, simply set is_function_calling_model=True and remove the tool_choice="none" override. The same code will then automatically use the richer tool‑calling flow.