Spaces:
Running
Running
Upload folder using huggingface_hub
Browse files- src/llm.py +44 -6
src/llm.py
CHANGED
|
@@ -20,6 +20,9 @@ logger = logging.getLogger(__name__)
|
|
| 20 |
# ── HuggingFace Inference API ─────────────────────────────
|
| 21 |
_hf_client = None
|
| 22 |
|
|
|
|
|
|
|
|
|
|
| 23 |
def _init_hf():
|
| 24 |
global _hf_client
|
| 25 |
token = os.getenv("HF_TOKEN")
|
|
@@ -38,7 +41,25 @@ def _init_hf():
|
|
| 38 |
logger.error(f"HF Inference API init failed: {e}")
|
| 39 |
return False
|
| 40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
# ── Groq fallback (works locally, may be blocked on HF Spaces) ──
|
|
|
|
| 42 |
_groq_client = None
|
| 43 |
|
| 44 |
def _init_groq():
|
|
@@ -69,14 +90,31 @@ def _call_hf(messages: list) -> str:
|
|
| 69 |
)
|
| 70 |
return response.choices[0].message.content
|
| 71 |
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
model="llama-3.3-70b-versatile",
|
| 77 |
messages=messages,
|
|
|
|
| 78 |
temperature=0.3,
|
| 79 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
)
|
| 81 |
return response.choices[0].message.content
|
| 82 |
|
|
|
|
| 20 |
# ── HuggingFace Inference API ─────────────────────────────
|
| 21 |
_hf_client = None
|
| 22 |
|
| 23 |
+
# ── OpenRouter (free tier, reliable fallback) ──────────────
|
| 24 |
+
_openrouter_client = None
|
| 25 |
+
|
| 26 |
def _init_hf():
|
| 27 |
global _hf_client
|
| 28 |
token = os.getenv("HF_TOKEN")
|
|
|
|
| 41 |
logger.error(f"HF Inference API init failed: {e}")
|
| 42 |
return False
|
| 43 |
|
| 44 |
+
def _init_openrouter():
|
| 45 |
+
global _openrouter_client
|
| 46 |
+
api_key = os.getenv("OPENROUTER_API_KEY")
|
| 47 |
+
if not api_key:
|
| 48 |
+
return False
|
| 49 |
+
try:
|
| 50 |
+
from openai import OpenAI
|
| 51 |
+
_openrouter_client = OpenAI(
|
| 52 |
+
base_url="https://openrouter.ai/api/v1",
|
| 53 |
+
api_key=api_key,
|
| 54 |
+
)
|
| 55 |
+
logger.info("OpenRouter ready as fallback")
|
| 56 |
+
return True
|
| 57 |
+
except Exception as e:
|
| 58 |
+
logger.error(f"OpenRouter init failed: {e}")
|
| 59 |
+
return False
|
| 60 |
+
|
| 61 |
# ── Groq fallback (works locally, may be blocked on HF Spaces) ──
|
| 62 |
+
_openrouter_ready = _init_openrouter()
|
| 63 |
_groq_client = None
|
| 64 |
|
| 65 |
def _init_groq():
|
|
|
|
| 90 |
)
|
| 91 |
return response.choices[0].message.content
|
| 92 |
|
| 93 |
+
openrouter(messages: list) -> str:
|
| 94 |
+
"""Call OpenRouter free tier."""
|
| 95 |
+
response = _openrouter_client.chat.completions.create(
|
| 96 |
+
model="meta-llama/llama-3.3-70b-instruct:free",
|
|
|
|
| 97 |
messages=messages,
|
| 98 |
+
max_tokens=1500,
|
| 99 |
temperature=0.3,
|
| 100 |
+
)
|
| 101 |
+
return response.choices[0].message.content
|
| 102 |
+
|
| 103 |
+
|
| 104 |
+
def _call_
|
| 105 |
+
def _call_groq(messages: list) -> str:
|
| 106 |
+
"""Call Groq as fthen OpenRouter, then Groq."""
|
| 107 |
+
if _hf_ready and _hf_client:
|
| 108 |
+
try:
|
| 109 |
+
return _call_hf(messages)
|
| 110 |
+
except Exception as e:
|
| 111 |
+
logger.warning(f"HF Inference failed: {e}, trying OpenRouter")
|
| 112 |
+
|
| 113 |
+
if _openrouter_ready and _openrouter_client:
|
| 114 |
+
try:
|
| 115 |
+
return _call_openrouter(messages)
|
| 116 |
+
except Exception as e:
|
| 117 |
+
logger.warning(f"OpenRouter
|
| 118 |
)
|
| 119 |
return response.choices[0].message.content
|
| 120 |
|