Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
"""Document parts map to Anthropic document blocks and OpenAI input_file; unknown ones are dropped."""
import asyncio
import json
import httpx
from core.inference import external_provider as ep_mod
from core.inference.external_provider import ExternalProviderClient
def _drive(coro):
return asyncio.new_event_loop().run_until_complete(coro)
def _capture(monkeypatch, *, provider: str, base_url: str, messages) -> dict:
captured: dict = {}
def handler(request: httpx.Request) -> httpx.Response:
captured["body"] = json.loads(request.content.decode("utf-8"))
if provider == "anthropic":
body = b"event: message_stop\n" b'data: {"type": "message_stop"}\n\n'
else:
body = (
b"event: response.completed\n"
b'data: {"type":"response.completed",'
b'"response":{"output":[],"usage":{"input_tokens":0,'
b'"output_tokens":0}}}\n\n'
)
return httpx.Response(
200,
content = body,
headers = {"content-type": "text/event-stream"},
)
monkeypatch.setattr(
ep_mod,
"_http_client",
httpx.AsyncClient(transport = httpx.MockTransport(handler)),
)
async def run():
client = ExternalProviderClient(
provider_type = provider,
base_url = base_url,
api_key = "sk-test",
)
kwargs = {
"messages": messages,
"model": "claude-opus-4-7" if provider == "anthropic" else "gpt-5.5",
"temperature": 0.7,
"top_p": 0.95,
"max_tokens": 32,
}
if provider == "openai":
kwargs["reasoning_effort"] = "medium"
async for _ in client.stream_chat_completion(**kwargs):
pass
await client.close()
_drive(run())
return captured
_TINY_PDF_B64 = "JVBERi0xLjQKJcOkw7zDtsOfCjEgMCBvYmoKPDw+PgplbmRvYmoK"
_PDF_DATA_URI = f"data:application/pdf;base64,{_TINY_PDF_B64}"
def _strip_cache(p: dict) -> dict:
return {k: v for k, v in p.items() if k != "cache_control"}
def test_anthropic_base64_pdf_becomes_document_block(monkeypatch):
captured = _capture(
monkeypatch,
provider = "anthropic",
base_url = "https://api.anthropic.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Summarise this paper."},
{
"type": "input_document",
"file_data": _PDF_DATA_URI,
"filename": "paper.pdf",
},
],
}
],
)
user_msg = captured["body"]["messages"][0]
parts = user_msg["content"]
types = [p.get("type") for p in parts]
assert "document" in types, parts
doc = _strip_cache(next(p for p in parts if p.get("type") == "document"))
assert doc == {
"type": "document",
"source": {
"type": "base64",
"media_type": "application/pdf",
"data": _TINY_PDF_B64,
},
"citations": {"enabled": True},
"title": "paper.pdf",
}
def test_anthropic_url_pdf_becomes_document_block(monkeypatch):
captured = _capture(
monkeypatch,
provider = "anthropic",
base_url = "https://api.anthropic.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Read this URL."},
{
"type": "input_document",
"file_url": "https://example.com/doc.pdf",
},
],
}
],
)
parts = captured["body"]["messages"][0]["content"]
doc = _strip_cache(next(p for p in parts if p.get("type") == "document"))
assert doc == {
"type": "document",
"source": {"type": "url", "url": "https://example.com/doc.pdf"},
"citations": {"enabled": True},
}
def test_anthropic_empty_document_part_is_dropped(monkeypatch):
captured = _capture(
monkeypatch,
provider = "anthropic",
base_url = "https://api.anthropic.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Hi."},
{"type": "input_document"},
],
}
],
)
parts = captured["body"]["messages"][0]["content"]
types = [p.get("type") for p in parts]
assert "document" not in types, parts
def test_anthropic_empty_only_document_drops_whole_message(monkeypatch):
# Anthropic 400s on an empty-content message.
captured = _capture(
monkeypatch,
provider = "anthropic",
base_url = "https://api.anthropic.com/v1",
messages = [
{"role": "user", "content": [{"type": "input_document"}]},
{"role": "user", "content": "but THIS one is fine"},
],
)
msgs = captured["body"]["messages"]
assert len(msgs) == 1, msgs
def test_anthropic_empty_data_uri_payload_is_dropped(monkeypatch):
captured = _capture(
monkeypatch,
provider = "anthropic",
base_url = "https://api.anthropic.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "still here"},
{
"type": "input_document",
"file_data": "data:application/pdf;base64,",
"filename": "empty.pdf",
},
{
"type": "input_document",
"file_data": "data:application/pdf;base64, ",
"filename": "whitespace.pdf",
},
],
}
],
)
parts = captured["body"]["messages"][0]["content"]
assert all(p.get("type") != "document" for p in parts), parts
def test_anthropic_empty_data_uri_falls_back_to_file_url(monkeypatch):
captured = _capture(
monkeypatch,
provider = "anthropic",
base_url = "https://api.anthropic.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Read this."},
{
"type": "input_document",
"file_data": "data:application/pdf;base64,",
"file_url": "https://example.com/doc.pdf",
"filename": "doc.pdf",
},
],
}
],
)
parts = captured["body"]["messages"][0]["content"]
doc = _strip_cache(next(p for p in parts if p.get("type") == "document"))
assert doc == {
"type": "document",
"source": {"type": "url", "url": "https://example.com/doc.pdf"},
"citations": {"enabled": True},
"title": "doc.pdf",
}
def test_anthropic_whitespace_only_data_uri_falls_back_to_file_url(monkeypatch):
captured = _capture(
monkeypatch,
provider = "anthropic",
base_url = "https://api.anthropic.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Read this."},
{
"type": "input_document",
"file_data": "data:application/pdf;base64, ",
"file_url": "https://example.com/doc.pdf",
},
],
}
],
)
parts = captured["body"]["messages"][0]["content"]
doc = _strip_cache(next(p for p in parts if p.get("type") == "document"))
assert doc == {
"type": "document",
"source": {"type": "url", "url": "https://example.com/doc.pdf"},
"citations": {"enabled": True},
}
def test_openai_base64_pdf_becomes_input_file(monkeypatch):
captured = _capture(
monkeypatch,
provider = "openai",
base_url = "https://api.openai.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Summarise this paper."},
{
"type": "input_document",
"file_data": _PDF_DATA_URI,
"filename": "paper.pdf",
},
],
}
],
)
user_msg = captured["body"]["input"][0]
parts = user_msg["content"]
fileblk = next(p for p in parts if p.get("type") == "input_file")
assert fileblk == {"type": "input_file", "file_data": _PDF_DATA_URI, "filename": "paper.pdf"}
def test_openai_url_pdf_becomes_input_file(monkeypatch):
captured = _capture(
monkeypatch,
provider = "openai",
base_url = "https://api.openai.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Read this URL."},
{
"type": "input_document",
"file_url": "https://example.com/doc.pdf",
},
],
}
],
)
parts = captured["body"]["input"][0]["content"]
fileblk = next(p for p in parts if p.get("type") == "input_file")
assert fileblk == {"type": "input_file", "file_url": "https://example.com/doc.pdf"}
def test_openai_empty_data_uri_falls_back_to_file_url(monkeypatch):
captured = _capture(
monkeypatch,
provider = "openai",
base_url = "https://api.openai.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Read this."},
{
"type": "input_document",
"file_data": "data:application/pdf;base64,",
"file_url": "https://example.com/doc.pdf",
"filename": "doc.pdf",
},
],
}
],
)
parts = captured["body"]["input"][0]["content"]
fileblk = next(p for p in parts if p.get("type") == "input_file")
assert "file_data" not in fileblk, fileblk
assert fileblk["file_url"] == "https://example.com/doc.pdf"
assert fileblk["filename"] == "doc.pdf"
def test_openai_whitespace_only_data_uri_falls_back_to_file_url(monkeypatch):
captured = _capture(
monkeypatch,
provider = "openai",
base_url = "https://api.openai.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Read this."},
{
"type": "input_document",
"file_data": "data:application/pdf;base64, ",
"file_url": "https://example.com/doc.pdf",
},
],
}
],
)
parts = captured["body"]["input"][0]["content"]
fileblk = next(p for p in parts if p.get("type") == "input_file")
assert "file_data" not in fileblk, fileblk
assert fileblk["file_url"] == "https://example.com/doc.pdf"
def test_openai_empty_data_uri_without_fallback_is_dropped(monkeypatch):
captured = _capture(
monkeypatch,
provider = "openai",
base_url = "https://api.openai.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Hi."},
{
"type": "input_document",
"file_data": "data:application/pdf;base64,",
"filename": "empty.pdf",
},
],
}
],
)
parts = captured["body"]["input"][0]["content"]
types = [p.get("type") for p in parts]
assert "input_file" not in types, parts
def test_openai_empty_document_part_is_dropped(monkeypatch):
captured = _capture(
monkeypatch,
provider = "openai",
base_url = "https://api.openai.com/v1",
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Hi."},
{"type": "input_document"},
],
}
],
)
parts = captured["body"]["input"][0]["content"]
types = [p.get("type") for p in parts]
assert "input_file" not in types, parts
def test_chat_message_accepts_input_document_part():
from models.inference import ChatMessage
msg = ChatMessage.model_validate(
{
"role": "user",
"content": [
{"type": "text", "text": "look"},
{
"type": "input_document",
"file_data": _PDF_DATA_URI,
"filename": "paper.pdf",
"media_type": "application/pdf",
},
],
}
)
assert isinstance(msg.content, list)
assert msg.content[1].type == "input_document"
assert msg.content[1].file_data == _PDF_DATA_URI
assert msg.content[1].filename == "paper.pdf"
assert msg.content[1].media_type == "application/pdf"
def test_build_external_messages_passes_input_document_for_native_and_custom_responses():
from models.inference import ChatMessage
from routes.inference import _build_external_messages
msgs = [
ChatMessage.model_validate(
{
"role": "user",
"content": [
{"type": "text", "text": "summarise"},
{
"type": "input_document",
"file_url": "https://example.com/doc.pdf",
"filename": "doc.pdf",
},
],
}
)
]
for provider, api_type in (("anthropic", None), ("openai", None), ("custom", "responses")):
out = _build_external_messages(
msgs, supports_vision = True, provider_type = provider, api_type = api_type
)
assert len(out) == 1, (provider, out)
parts = out[0]["content"]
assert parts[0] == {"type": "text", "text": "summarise"}, provider
assert parts[1] == {
"type": "input_document",
"file_url": "https://example.com/doc.pdf",
"filename": "doc.pdf",
}, provider
def test_build_external_messages_strips_input_document_for_unmapped_providers():
# Generic /chat/completions passthrough providers forward messages verbatim; strip input_document.
from models.inference import ChatMessage
from routes.inference import _build_external_messages
msgs = [
ChatMessage.model_validate(
{
"role": "user",
"content": [
{"type": "text", "text": "summarise"},
{
"type": "input_document",
"file_url": "https://example.com/doc.pdf",
"filename": "doc.pdf",
},
],
}
)
]
for provider in ("gemini", "mistral", "kimi", "openrouter", "deepseek", "qwen", "custom"):
api_type = "chat_completions" if provider == "custom" else None
out = _build_external_messages(
msgs, supports_vision = True, provider_type = provider, api_type = api_type
)
assert len(out) == 1, (provider, out)
parts = out[0]["content"]
types = [p.get("type") for p in parts if isinstance(p, dict)]
assert "input_document" not in types, (provider, parts)
assert {"type": "text", "text": "summarise"} in parts, (provider, parts)
def test_build_external_messages_strips_input_document_when_provider_type_unknown():
from models.inference import ChatMessage
from routes.inference import _build_external_messages
msgs = [
ChatMessage.model_validate(
{
"role": "user",
"content": [
{"type": "text", "text": "summarise"},
{
"type": "input_document",
"file_data": _PDF_DATA_URI,
},
],
}
)
]
out = _build_external_messages(msgs, supports_vision = True)
parts = out[0]["content"]
types = [p.get("type") for p in parts if isinstance(p, dict)]
assert "input_document" not in types, parts
def test_build_external_messages_drops_input_document_for_non_vision_provider():
from models.inference import ChatMessage
from routes.inference import _build_external_messages
msgs = [
ChatMessage.model_validate(
{
"role": "user",
"content": [
{"type": "text", "text": "summarise"},
{
"type": "input_document",
"file_data": _PDF_DATA_URI,
},
],
}
)
]
out = _build_external_messages(msgs, supports_vision = False)
assert out == [{"role": "user", "content": "summarise"}]