DeepSeek Provider
DeepSeek provider implementation for the AiCore LLM system.
Features
- Supports DeepSeek's chat and reasoning models
- Implements token counting and cost tracking
- Includes happy hour pricing support
- Compatible with OpenAI API format
- Full observability integration (metrics, logging, tracing)
Configuration
python
from aicore.llm.config import LlmConfig
config = LlmConfig(
provider="deepseek",
api_key="your_api_key", # Get from DeepSeek console
model="deepseek-chat", # or "deepseek-reasoner"
base_url="https://api.deepseek.com/v1", # Optional custom endpoint
timeout=30, # Request timeout in seconds
max_retries=3 # Automatic retry attempts
)Supported Models
| Model Name | Description | Context Window |
|---|---|---|
deepseek-chat | General purpose chat model | 64K tokens |
deepseek-reasoner | Specialized reasoning model | 64K tokens |
Usage Example
python
from aicore.llm import Llm
# Initialize with config
llm = Llm(config=config)
# Basic completion
response = llm.complete("Explain quantum computing in simple terms")Observability Integration
All DeepSeek operations automatically track:
- Token usage (input/output)
- Latency metrics
- Cost calculations
- Success/failure rates
Access metrics through the Observability Dashboard.
Pricing
DeepSeek uses token-based pricing. Current rates can be accessed programmatically:
python
from aicore.models_metadata import METADATA
model_data = METADATA["deepseek-deepseek-chat"]
print(f"Input: ${model_data.pricing.input_per_token:.8f}/token")
print(f"Output: ${model_data.pricing.output_per_token:.8f}/token")Error Handling
The provider implements automatic retries for:
- Rate limits (429 errors)
- Server errors (5xx)
- Temporary network issues
Customize retry behavior in the LlmConfig.