Spaces:
Running
Running
File size: 4,818 Bytes
8dcf472 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 | """
DealFlow AI — Configuration
Loads env vars and constructs the LLM clients for CrewAI.
AMD Fallback Status: LOCKED at T+63h (2026-05-06).
Primary backend: HuggingFace Serverless Inference API (Qwen/Qwen2.5-72B-Instruct).
AMD MI300X bonus track abandoned per PRE-FLIGHT SPEC #3.
"""
from __future__ import annotations
import os
from enum import Enum
from pathlib import Path
from typing import Optional
from dotenv import load_dotenv
from loguru import logger
from pydantic import BaseModel, Field
load_dotenv()
class Backend(str, Enum):
VLLM = "vllm"
HF = "hf"
OPENAI = "openai"
# HF Serverless Inference — featherless-ai provider (supports Qwen2.5-72B)
# Verified working endpoint for Qwen/Qwen2.5-72B-Instruct
HF_ROUTER_BASE_URL = "https://router.huggingface.co/featherless-ai/v1"
def hf_base_url_for(model: str) -> str: # noqa: ARG001 — kept for future provider routing
return HF_ROUTER_BASE_URL
# Primary model locked at T+63h (AMD credits never arrived)
HF_PRIMARY_MODEL = "Qwen/Qwen2.5-72B-Instruct"
HF_FALLBACK_MODEL = "Qwen/Qwen2.5-32B-Instruct" # if 72B latency is an issue
class AppConfig(BaseModel):
# Inference backend — default is now "hf" (AMD locked out at T+63h)
llm_backend: Backend = Field(
default_factory=lambda: Backend(os.getenv("LLM_BACKEND", "hf"))
)
# HuggingFace Serverless (primary)
hf_token: str = Field(
default_factory=lambda: (
os.environ.get("HF_TOKEN")
or os.environ.get("HUGGINGFACE_TOKEN")
or ""
)
)
hf_model: str = Field(
default_factory=lambda: os.getenv("HF_MODEL", HF_PRIMARY_MODEL)
)
# Computed dynamically from hf_model if not explicitly overridden
hf_base_url: Optional[str] = Field(
default_factory=lambda: os.getenv("HF_BASE_URL")
)
# vLLM (AMD MI300X — kept for future use, currently inactive)
vllm_base_url: str = Field(
default_factory=lambda: os.getenv("VLLM_BASE_URL", "http://localhost:8000/v1")
)
vllm_api_key: str = Field(
default_factory=lambda: os.environ.get("VLLM_API_KEY", "")
)
vllm_model: str = Field(
default_factory=lambda: os.getenv("VLLM_MODEL", "Qwen/Qwen3-VL-32B-Instruct-FP8")
)
# Search
serper_api_key: Optional[str] = Field(
default_factory=lambda: os.environ.get("SERPER_API_KEY")
)
# Memory
mem0_api_key: Optional[str] = Field(
default_factory=lambda: os.getenv("MEM0_API_KEY")
)
mem0_base_url: Optional[str] = Field(
default_factory=lambda: os.getenv("MEM0_BASE_URL")
)
# App
output_dir: Path = Field(
default_factory=lambda: Path(os.getenv("OUTPUT_DIR", "./outputs"))
)
max_crew_iterations: int = Field(
default_factory=lambda: int(os.getenv("MAX_CREW_ITERATIONS", "3"))
)
verbose_agents: bool = Field(
default_factory=lambda: os.getenv("VERBOSE_AGENTS", "false").lower() == "true"
)
model_config = {"arbitrary_types_allowed": True}
def get_config() -> AppConfig:
return AppConfig()
def get_llm(config: Optional[AppConfig] = None):
"""Return a CrewAI-compatible LLM instance based on active backend."""
if config is None:
config = get_config()
from crewai import LLM
if config.llm_backend == Backend.HF:
if not config.hf_token:
raise ValueError("HF_TOKEN (or HUGGINGFACE_TOKEN) must be set for HF backend")
# HF Serverless uses model-specific OpenAI-compatible paths:
# https://api-inference.huggingface.co/models/{model}/v1/chat/completions
base_url = config.hf_base_url or hf_base_url_for(config.hf_model)
logger.info(
f"Using HF Serverless Inference: model={config.hf_model} base_url={base_url}"
)
return LLM(
model=f"openai/{config.hf_model}",
base_url=base_url,
api_key=config.hf_token,
temperature=0.1,
max_tokens=4096,
)
elif config.llm_backend == Backend.VLLM:
logger.info(
f"Using vLLM backend: {config.vllm_base_url} model={config.vllm_model}"
)
return LLM(
model=f"openai/{config.vllm_model}",
base_url=config.vllm_base_url,
api_key=config.vllm_api_key,
temperature=0.1,
max_tokens=4096,
)
elif config.llm_backend == Backend.OPENAI:
api_key = os.getenv("OPENAI_API_KEY")
if not api_key:
raise ValueError("OPENAI_API_KEY not set for openai backend")
return LLM(
model="gpt-4o",
api_key=api_key,
temperature=0.1,
max_tokens=4096,
)
else:
raise ValueError(f"Unknown LLM_BACKEND: {config.llm_backend}")
|