6cab9e56fc
fix: model definition re-introduced into the schema but intelligently and only a summary is generated per tool. Required to ensure CLI calls and uses the correct model fix: removed `model` param from some tools where this wasn't needed fix: fixed adherence to `*_ALLOWED_MODELS` by advertising only the allowed models to the CLI fix: removed duplicates across providers when passing canonical names back to the CLI; the first enabled provider wins
161 lines
6.3 KiB
Python
161 lines
6.3 KiB
Python
"""X.AI (GROK) model provider implementation."""
|
|
|
|
import logging
|
|
from typing import TYPE_CHECKING, Optional
|
|
|
|
if TYPE_CHECKING:
|
|
from tools.models import ToolModelCategory
|
|
|
|
from .openai_compatible import OpenAICompatibleProvider
|
|
from .shared import ModelCapabilities, ModelResponse, ProviderType, TemperatureConstraint
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
class XAIModelProvider(OpenAICompatibleProvider):
|
|
"""Integration for X.AI's GROK models exposed over an OpenAI-style API.
|
|
|
|
Publishes capability metadata for the officially supported deployments and
|
|
maps tool-category preferences to the appropriate GROK model.
|
|
"""
|
|
|
|
FRIENDLY_NAME = "X.AI"
|
|
|
|
# Model configurations using ModelCapabilities objects
|
|
MODEL_CAPABILITIES = {
|
|
"grok-4": ModelCapabilities(
|
|
provider=ProviderType.XAI,
|
|
model_name="grok-4",
|
|
friendly_name="X.AI (Grok 4)",
|
|
intelligence_score=16,
|
|
context_window=256_000, # 256K tokens
|
|
max_output_tokens=256_000, # 256K tokens max output
|
|
supports_extended_thinking=True, # Grok-4 supports reasoning mode
|
|
supports_system_prompts=True,
|
|
supports_streaming=True,
|
|
supports_function_calling=True, # Function calling supported
|
|
supports_json_mode=True, # Structured outputs supported
|
|
supports_images=True, # Multimodal capabilities
|
|
max_image_size_mb=20.0, # Standard image size limit
|
|
supports_temperature=True,
|
|
temperature_constraint=TemperatureConstraint.create("range"),
|
|
description="GROK-4 (256K context) - Frontier multimodal reasoning model with advanced capabilities",
|
|
aliases=["grok", "grok4", "grok-4"],
|
|
),
|
|
"grok-3": ModelCapabilities(
|
|
provider=ProviderType.XAI,
|
|
model_name="grok-3",
|
|
friendly_name="X.AI (Grok 3)",
|
|
intelligence_score=13,
|
|
context_window=131_072, # 131K tokens
|
|
max_output_tokens=131072,
|
|
supports_extended_thinking=False,
|
|
supports_system_prompts=True,
|
|
supports_streaming=True,
|
|
supports_function_calling=True,
|
|
supports_json_mode=False, # Assuming GROK doesn't have JSON mode yet
|
|
supports_images=False, # Assuming GROK is text-only for now
|
|
max_image_size_mb=0.0,
|
|
supports_temperature=True,
|
|
temperature_constraint=TemperatureConstraint.create("range"),
|
|
description="GROK-3 (131K context) - Advanced reasoning model from X.AI, excellent for complex analysis",
|
|
aliases=["grok3"],
|
|
),
|
|
"grok-3-fast": ModelCapabilities(
|
|
provider=ProviderType.XAI,
|
|
model_name="grok-3-fast",
|
|
friendly_name="X.AI (Grok 3 Fast)",
|
|
intelligence_score=12,
|
|
context_window=131_072, # 131K tokens
|
|
max_output_tokens=131072,
|
|
supports_extended_thinking=False,
|
|
supports_system_prompts=True,
|
|
supports_streaming=True,
|
|
supports_function_calling=True,
|
|
supports_json_mode=False, # Assuming GROK doesn't have JSON mode yet
|
|
supports_images=False, # Assuming GROK is text-only for now
|
|
max_image_size_mb=0.0,
|
|
supports_temperature=True,
|
|
temperature_constraint=TemperatureConstraint.create("range"),
|
|
description="GROK-3 Fast (131K context) - Higher performance variant, faster processing but more expensive",
|
|
aliases=["grok3fast", "grokfast", "grok3-fast"],
|
|
),
|
|
}
|
|
|
|
def __init__(self, api_key: str, **kwargs):
|
|
"""Initialize X.AI provider with API key."""
|
|
# Set X.AI base URL
|
|
kwargs.setdefault("base_url", "https://api.x.ai/v1")
|
|
super().__init__(api_key, **kwargs)
|
|
|
|
def get_provider_type(self) -> ProviderType:
|
|
"""Get the provider type."""
|
|
return ProviderType.XAI
|
|
|
|
def generate_content(
|
|
self,
|
|
prompt: str,
|
|
model_name: str,
|
|
system_prompt: Optional[str] = None,
|
|
temperature: float = 0.3,
|
|
max_output_tokens: Optional[int] = None,
|
|
**kwargs,
|
|
) -> ModelResponse:
|
|
"""Generate content using X.AI API with proper model name resolution."""
|
|
# Resolve model alias before making API call
|
|
resolved_model_name = self._resolve_model_name(model_name)
|
|
|
|
# Call parent implementation with resolved model name
|
|
return super().generate_content(
|
|
prompt=prompt,
|
|
model_name=resolved_model_name,
|
|
system_prompt=system_prompt,
|
|
temperature=temperature,
|
|
max_output_tokens=max_output_tokens,
|
|
**kwargs,
|
|
)
|
|
|
|
def get_preferred_model(self, category: "ToolModelCategory", allowed_models: list[str]) -> Optional[str]:
|
|
"""Get XAI's preferred model for a given category from allowed models.
|
|
|
|
Args:
|
|
category: The tool category requiring a model
|
|
allowed_models: Pre-filtered list of models allowed by restrictions
|
|
|
|
Returns:
|
|
Preferred model name or None
|
|
"""
|
|
from tools.models import ToolModelCategory
|
|
|
|
if not allowed_models:
|
|
return None
|
|
|
|
if category == ToolModelCategory.EXTENDED_REASONING:
|
|
# Prefer GROK-4 for advanced reasoning with thinking mode
|
|
if "grok-4" in allowed_models:
|
|
return "grok-4"
|
|
elif "grok-3" in allowed_models:
|
|
return "grok-3"
|
|
# Fall back to any available model
|
|
return allowed_models[0]
|
|
|
|
elif category == ToolModelCategory.FAST_RESPONSE:
|
|
# Prefer GROK-3-Fast for speed, then GROK-4
|
|
if "grok-3-fast" in allowed_models:
|
|
return "grok-3-fast"
|
|
elif "grok-4" in allowed_models:
|
|
return "grok-4"
|
|
# Fall back to any available model
|
|
return allowed_models[0]
|
|
|
|
else: # BALANCED or default
|
|
# Prefer GROK-4 for balanced use (best overall capabilities)
|
|
if "grok-4" in allowed_models:
|
|
return "grok-4"
|
|
elif "grok-3" in allowed_models:
|
|
return "grok-3"
|
|
elif "grok-3-fast" in allowed_models:
|
|
return "grok-3-fast"
|
|
# Fall back to any available model
|
|
return allowed_models[0]
|