feat(providers): first-class Mistral support
Mistral's API constrains reasoning_effort to "high"/"none", rejects the
kwarg entirely for Magistral (reasoning is implicit), returns assistant
content as a mixed array of {type:"thinking",...}/{type:"text",...}
blocks, and 400s on the reasoning_content key in history.
- Remap user-supplied reasoning_effort (low/medium/minimal) onto Mistral's
two-tier vocabulary; strip the kwarg for Magistral models
- Lift thinking blocks into reasoning_content for both batch and streaming
responses; pass only text through on_content_delta callbacks
- Drop reasoning_content from outbound history when the spec asks for it
- Expose per-preset reasoning_effort_values so the UI can render the
provider-specific option set
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Javad Arjmandi
committed by
Xubin Ren
Xubin Ren
co-authored by
Claude Sonnet 4.6
Claude Sonnet 4.6
parent
51bd3337ef
commit
d5f5eb43e5
@@ -1,16 +1,32 @@
|
||||
"""Tests for the Mistral provider registration."""
|
||||
"""Tests for the Mistral provider registration and reasoning quirks."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from unittest.mock import patch
|
||||
|
||||
from nanobot.config.schema import ProvidersConfig
|
||||
from nanobot.providers.registry import PROVIDERS
|
||||
from nanobot.providers.openai_compat_provider import OpenAICompatProvider
|
||||
from nanobot.providers.registry import PROVIDERS, find_by_name
|
||||
|
||||
|
||||
def test_mistral_config_field_exists():
|
||||
def _mistral_provider(default_model: str = "mistral-medium-3-5") -> OpenAICompatProvider:
|
||||
spec = find_by_name("mistral")
|
||||
assert spec is not None
|
||||
with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
|
||||
return OpenAICompatProvider(
|
||||
api_key="test-key",
|
||||
default_model=default_model,
|
||||
spec=spec,
|
||||
)
|
||||
|
||||
|
||||
def test_mistral_config_field_exists() -> None:
|
||||
"""ProvidersConfig should have a mistral field."""
|
||||
config = ProvidersConfig()
|
||||
assert hasattr(config, "mistral")
|
||||
|
||||
|
||||
def test_mistral_provider_in_registry():
|
||||
def test_mistral_provider_in_registry() -> None:
|
||||
"""Mistral should be registered in the provider registry."""
|
||||
specs = {s.name: s for s in PROVIDERS}
|
||||
assert "mistral" in specs
|
||||
@@ -18,3 +34,322 @@ def test_mistral_provider_in_registry():
|
||||
mistral = specs["mistral"]
|
||||
assert mistral.env_key == "MISTRAL_API_KEY"
|
||||
assert mistral.default_api_base == "https://api.mistral.ai/v1"
|
||||
|
||||
|
||||
def test_mistral_keyword_match_covers_model_families() -> None:
|
||||
"""Codestral, Devstral, Ministral, Magistral models route to the Mistral spec."""
|
||||
from nanobot.config.schema import Config
|
||||
|
||||
for model in (
|
||||
"mistral-large-latest",
|
||||
"magistral-medium-latest",
|
||||
"ministral-8b-latest",
|
||||
"codestral-latest",
|
||||
"devstral-medium-latest",
|
||||
):
|
||||
config = Config.model_validate({
|
||||
"providers": {"mistral": {"apiKey": "test-key"}},
|
||||
"agents": {"defaults": {"model": model}},
|
||||
})
|
||||
assert config.get_provider_name(model) == "mistral", model
|
||||
|
||||
|
||||
def test_reasoning_effort_low_remaps_to_none_omitted() -> None:
|
||||
"""Mistral rejects low/medium efforts: low should map to "none" (omitted)."""
|
||||
p = _mistral_provider()
|
||||
kwargs = p._build_kwargs(
|
||||
messages=[{"role": "user", "content": "hi"}],
|
||||
tools=None,
|
||||
model="mistral-medium-3-5",
|
||||
max_tokens=64,
|
||||
temperature=0.5,
|
||||
reasoning_effort="low",
|
||||
tool_choice=None,
|
||||
)
|
||||
assert "reasoning_effort" not in kwargs
|
||||
|
||||
|
||||
def test_reasoning_effort_minimal_remaps_to_none_omitted() -> None:
|
||||
p = _mistral_provider()
|
||||
kwargs = p._build_kwargs(
|
||||
messages=[{"role": "user", "content": "hi"}],
|
||||
tools=None,
|
||||
model="mistral-vibe-cli-latest",
|
||||
max_tokens=64,
|
||||
temperature=0.5,
|
||||
reasoning_effort="minimal",
|
||||
tool_choice=None,
|
||||
)
|
||||
assert "reasoning_effort" not in kwargs
|
||||
|
||||
|
||||
def test_reasoning_effort_medium_remaps_to_high() -> None:
|
||||
"""Mistral has no 'medium' tier: bump up to 'high'."""
|
||||
p = _mistral_provider()
|
||||
kwargs = p._build_kwargs(
|
||||
messages=[{"role": "user", "content": "hi"}],
|
||||
tools=None,
|
||||
model="mistral-medium-3-5",
|
||||
max_tokens=64,
|
||||
temperature=0.5,
|
||||
reasoning_effort="medium",
|
||||
tool_choice=None,
|
||||
)
|
||||
assert kwargs["reasoning_effort"] == "high"
|
||||
|
||||
|
||||
def test_reasoning_effort_high_passes_through() -> None:
|
||||
p = _mistral_provider()
|
||||
kwargs = p._build_kwargs(
|
||||
messages=[{"role": "user", "content": "hi"}],
|
||||
tools=None,
|
||||
model="mistral-medium-3-5",
|
||||
max_tokens=64,
|
||||
temperature=0.5,
|
||||
reasoning_effort="high",
|
||||
tool_choice=None,
|
||||
)
|
||||
assert kwargs["reasoning_effort"] == "high"
|
||||
|
||||
|
||||
def test_magistral_strips_reasoning_effort() -> None:
|
||||
"""Magistral reasons implicitly; API rejects reasoning_effort kwarg."""
|
||||
p = _mistral_provider()
|
||||
for effort in ("low", "medium", "high", "minimal"):
|
||||
kwargs = p._build_kwargs(
|
||||
messages=[{"role": "user", "content": "hi"}],
|
||||
tools=None,
|
||||
model="magistral-medium-latest",
|
||||
max_tokens=64,
|
||||
temperature=0.5,
|
||||
reasoning_effort=effort,
|
||||
tool_choice=None,
|
||||
)
|
||||
assert "reasoning_effort" not in kwargs, effort
|
||||
|
||||
|
||||
def test_extract_thinking_content_from_mistral_response() -> None:
|
||||
"""Thinking blocks should land in reasoning_content, not content."""
|
||||
p = _mistral_provider()
|
||||
response = {
|
||||
"choices": [
|
||||
{
|
||||
"finish_reason": "stop",
|
||||
"message": {
|
||||
"role": "assistant",
|
||||
"content": [
|
||||
{
|
||||
"type": "thinking",
|
||||
"thinking": [
|
||||
{"type": "text", "text": "Let me think..."}
|
||||
],
|
||||
"closed": True,
|
||||
},
|
||||
{"type": "text", "text": "Final answer is 35."},
|
||||
],
|
||||
},
|
||||
}
|
||||
],
|
||||
"usage": {"prompt_tokens": 10, "completion_tokens": 20, "total_tokens": 30},
|
||||
}
|
||||
parsed = p._parse(response)
|
||||
assert parsed.content == "Final answer is 35."
|
||||
assert parsed.reasoning_content == "Let me think..."
|
||||
|
||||
|
||||
def test_extract_thinking_content_with_tool_calls() -> None:
|
||||
"""A response with thinking + tool_calls (no text content) should still parse."""
|
||||
p = _mistral_provider()
|
||||
response = {
|
||||
"choices": [
|
||||
{
|
||||
"finish_reason": "tool_calls",
|
||||
"message": {
|
||||
"role": "assistant",
|
||||
"tool_calls": [
|
||||
{
|
||||
"id": "abc123def",
|
||||
"type": "function",
|
||||
"function": {
|
||||
"name": "get_weather",
|
||||
"arguments": '{"city": "Paris"}',
|
||||
},
|
||||
}
|
||||
],
|
||||
"content": [
|
||||
{
|
||||
"type": "thinking",
|
||||
"thinking": [
|
||||
{"type": "text", "text": "I should call get_weather."}
|
||||
],
|
||||
}
|
||||
],
|
||||
},
|
||||
}
|
||||
],
|
||||
}
|
||||
parsed = p._parse(response)
|
||||
assert parsed.content in (None, "")
|
||||
assert parsed.reasoning_content == "I should call get_weather."
|
||||
assert len(parsed.tool_calls) == 1
|
||||
assert parsed.tool_calls[0].name == "get_weather"
|
||||
assert parsed.tool_calls[0].arguments == {"city": "Paris"}
|
||||
|
||||
|
||||
def test_thinking_content_only_for_mistral_spec() -> None:
|
||||
"""Providers without extract_thinking_blocks should not lift thinking text."""
|
||||
other_spec = find_by_name("openai")
|
||||
assert other_spec is not None
|
||||
with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
|
||||
p = OpenAICompatProvider(
|
||||
api_key="test-key",
|
||||
default_model="gpt-4o",
|
||||
spec=other_spec,
|
||||
)
|
||||
|
||||
response = {
|
||||
"choices": [
|
||||
{
|
||||
"finish_reason": "stop",
|
||||
"message": {
|
||||
"role": "assistant",
|
||||
"content": [
|
||||
{
|
||||
"type": "thinking",
|
||||
"thinking": [{"type": "text", "text": "secret"}],
|
||||
},
|
||||
{"type": "text", "text": "hi"},
|
||||
],
|
||||
},
|
||||
}
|
||||
],
|
||||
}
|
||||
parsed = p._parse(response)
|
||||
assert parsed.content == "hi"
|
||||
assert parsed.reasoning_content is None
|
||||
|
||||
|
||||
def test_streaming_thinking_chunks_become_reasoning() -> None:
|
||||
"""Streamed thinking deltas (Mistral shape) feed reasoning_content."""
|
||||
chunks = [
|
||||
{
|
||||
"choices": [
|
||||
{
|
||||
"delta": {
|
||||
"content": [
|
||||
{
|
||||
"type": "thinking",
|
||||
"thinking": [{"type": "text", "text": "step 1 "}],
|
||||
}
|
||||
]
|
||||
},
|
||||
"finish_reason": None,
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"choices": [
|
||||
{
|
||||
"delta": {
|
||||
"content": [
|
||||
{
|
||||
"type": "thinking",
|
||||
"thinking": [{"type": "text", "text": "step 2"}],
|
||||
}
|
||||
]
|
||||
},
|
||||
"finish_reason": None,
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"choices": [
|
||||
{
|
||||
"delta": {"content": "final"},
|
||||
"finish_reason": "stop",
|
||||
}
|
||||
]
|
||||
},
|
||||
]
|
||||
parsed = OpenAICompatProvider._parse_chunks(chunks)
|
||||
assert parsed.content == "final"
|
||||
assert parsed.reasoning_content == "step 1 step 2"
|
||||
|
||||
|
||||
def test_streaming_thinking_chunks_via_sdk_path() -> None:
|
||||
"""The SDK-style branch must coerce list-shaped delta.content to text.
|
||||
|
||||
Regression: Mistral's vibe-cli/medium-3-5 streamed delta.content as a
|
||||
list, which was previously appended verbatim to content_parts and blew
|
||||
up later with "can only concatenate str (not list) to str".
|
||||
"""
|
||||
|
||||
class _Delta:
|
||||
def __init__(self, content):
|
||||
self.content = content
|
||||
self.tool_calls = None
|
||||
self.function_call = None
|
||||
|
||||
class _Choice:
|
||||
def __init__(self, content, finish=None):
|
||||
self.delta = _Delta(content)
|
||||
self.finish_reason = finish
|
||||
|
||||
class _Chunk:
|
||||
def __init__(self, content, finish=None):
|
||||
self.choices = [_Choice(content, finish)]
|
||||
|
||||
chunks = [
|
||||
_Chunk([{"type": "thinking", "thinking": [{"type": "text", "text": "ponder"}]}]),
|
||||
_Chunk([{"type": "text", "text": "Hello "}]),
|
||||
_Chunk("world.", finish="stop"),
|
||||
]
|
||||
parsed = OpenAICompatProvider._parse_chunks(chunks)
|
||||
assert parsed.content == "Hello world."
|
||||
assert parsed.reasoning_content == "ponder"
|
||||
|
||||
|
||||
def test_mistral_strips_reasoning_content_from_history() -> None:
|
||||
"""Mistral's request schema 400s on reasoning_content; it must be dropped."""
|
||||
p = _mistral_provider()
|
||||
messages = [
|
||||
{"role": "user", "content": "hi"},
|
||||
{
|
||||
"role": "assistant",
|
||||
"content": "Hello!",
|
||||
"reasoning_content": "internal thoughts",
|
||||
},
|
||||
{"role": "user", "content": "follow up"},
|
||||
]
|
||||
sanitized = p._sanitize_messages(messages)
|
||||
assert all("reasoning_content" not in msg for msg in sanitized)
|
||||
assert sanitized[1]["content"] == "Hello!"
|
||||
|
||||
|
||||
def test_mistral_tool_call_ids_get_normalized() -> None:
|
||||
"""Non-9-char tool_call IDs should be hashed to 9-char alphanumeric."""
|
||||
p = _mistral_provider()
|
||||
messages = [
|
||||
{"role": "user", "content": "hi"},
|
||||
{
|
||||
"role": "assistant",
|
||||
"tool_calls": [
|
||||
{
|
||||
"id": "call_abc_xyz_too_long_for_mistral",
|
||||
"type": "function",
|
||||
"function": {"name": "x", "arguments": "{}"},
|
||||
}
|
||||
],
|
||||
},
|
||||
{
|
||||
"role": "tool",
|
||||
"tool_call_id": "call_abc_xyz_too_long_for_mistral",
|
||||
"content": "ok",
|
||||
},
|
||||
]
|
||||
sanitized = p._sanitize_messages(messages)
|
||||
assistant_id = sanitized[1]["tool_calls"][0]["id"]
|
||||
tool_id = sanitized[2]["tool_call_id"]
|
||||
assert len(assistant_id) == 9
|
||||
assert assistant_id.isalnum()
|
||||
assert assistant_id == tool_id
|
||||
|
||||
Reference in New Issue
Block a user