Commit ·
f69f6eb
1
Parent(s): 4ff9aed
fix: DNS endpoint discovery - test multiple HF hostnames to find reachable one
Browse files- src/hf_inference.py +79 -36
src/hf_inference.py
CHANGED
|
@@ -1,21 +1,22 @@
|
|
| 1 |
"""
|
| 2 |
HuggingFace Inference API client for Bayan models.
|
| 3 |
|
| 4 |
-
|
| 5 |
-
|
| 6 |
"""
|
| 7 |
|
| 8 |
import os
|
| 9 |
import json
|
| 10 |
import logging
|
| 11 |
import time
|
|
|
|
| 12 |
|
| 13 |
logger = logging.getLogger(__name__)
|
| 14 |
|
| 15 |
HF_API_TOKEN = os.environ.get("HF_API_TOKEN", "")
|
| 16 |
-
HF_API_BASE = "https://api-inference.huggingface.co/models/"
|
| 17 |
|
| 18 |
_client = None
|
|
|
|
| 19 |
|
| 20 |
|
| 21 |
def _get_client():
|
|
@@ -33,45 +34,66 @@ PUNCTUATION_REPO = os.environ.get("PUNCTUATION_REPO_ID", "bayan10/PuncAra-v1")
|
|
| 33 |
AUTOCOMPLETE_REPO = os.environ.get("AUTOCOMPLETE_REPO_ID", "bayan10/AutoComplete")
|
| 34 |
|
| 35 |
|
| 36 |
-
def
|
| 37 |
-
"""
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
RequestParameters(url, task, model, json, data, headers)
|
| 41 |
-
"""
|
| 42 |
-
from huggingface_hub.inference._common import RequestParameters
|
| 43 |
|
| 44 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 45 |
|
| 46 |
if "options" not in payload:
|
| 47 |
payload["options"] = {"wait_for_model": True}
|
| 48 |
|
| 49 |
-
|
| 50 |
-
|
|
|
|
| 51 |
|
| 52 |
-
|
| 53 |
-
url=url,
|
| 54 |
-
task=task,
|
| 55 |
-
model=repo_id,
|
| 56 |
-
json=payload,
|
| 57 |
-
data=None,
|
| 58 |
-
headers={},
|
| 59 |
-
)
|
| 60 |
|
| 61 |
-
|
|
|
|
| 62 |
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
result = json.loads(response.decode("utf-8"))
|
| 66 |
-
elif isinstance(response, str):
|
| 67 |
-
result = json.loads(response)
|
| 68 |
-
else:
|
| 69 |
-
result = response
|
| 70 |
|
|
|
|
| 71 |
logger.info("HF result for %s: type=%s preview=%s",
|
| 72 |
repo_id, type(result).__name__, str(result)[:200])
|
| 73 |
|
| 74 |
-
# Check for API errors in response
|
| 75 |
if isinstance(result, dict) and "error" in result:
|
| 76 |
raise RuntimeError("HF API error: " + str(result["error"]))
|
| 77 |
|
|
@@ -101,7 +123,7 @@ def _extract_text(result, fallback=""):
|
|
| 101 |
# ============================================================
|
| 102 |
|
| 103 |
def hf_summarize(text, max_length=128, min_length=30):
|
| 104 |
-
result =
|
| 105 |
"inputs": text,
|
| 106 |
"parameters": {"max_length": max_length, "min_length": min_length},
|
| 107 |
}, task="summarization")
|
|
@@ -109,17 +131,17 @@ def hf_summarize(text, max_length=128, min_length=30):
|
|
| 109 |
|
| 110 |
|
| 111 |
def hf_correct_spelling(text):
|
| 112 |
-
result =
|
| 113 |
return _extract_text(result, text)
|
| 114 |
|
| 115 |
|
| 116 |
def hf_add_punctuation(text):
|
| 117 |
-
result =
|
| 118 |
return _extract_text(result, text)
|
| 119 |
|
| 120 |
|
| 121 |
def hf_autocomplete(text, n=5):
|
| 122 |
-
result =
|
| 123 |
"inputs": text,
|
| 124 |
"parameters": {"max_new_tokens": 20},
|
| 125 |
}, task="text-generation")
|
|
@@ -141,17 +163,38 @@ def hf_autocomplete(text, n=5):
|
|
| 141 |
|
| 142 |
def check_hf_api_available():
|
| 143 |
try:
|
| 144 |
-
return
|
| 145 |
except Exception:
|
| 146 |
return False
|
| 147 |
|
| 148 |
|
| 149 |
def debug_test_all_models():
|
| 150 |
-
"""Test all HF models."""
|
| 151 |
results = {}
|
| 152 |
test_text = "هذا نص تجريبي للاختبار"
|
| 153 |
long_text = (test_text + " ") * 5
|
| 154 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 155 |
try:
|
| 156 |
import huggingface_hub
|
| 157 |
results["_info"] = {"hf_hub_version": huggingface_hub.__version__}
|
|
|
|
| 1 |
"""
|
| 2 |
HuggingFace Inference API client for Bayan models.
|
| 3 |
|
| 4 |
+
Dynamically discovers a reachable HF API endpoint from inside HF Spaces,
|
| 5 |
+
since api-inference.huggingface.co is not DNS-resolvable from free-tier containers.
|
| 6 |
"""
|
| 7 |
|
| 8 |
import os
|
| 9 |
import json
|
| 10 |
import logging
|
| 11 |
import time
|
| 12 |
+
import socket
|
| 13 |
|
| 14 |
logger = logging.getLogger(__name__)
|
| 15 |
|
| 16 |
HF_API_TOKEN = os.environ.get("HF_API_TOKEN", "")
|
|
|
|
| 17 |
|
| 18 |
_client = None
|
| 19 |
+
_working_url = None
|
| 20 |
|
| 21 |
|
| 22 |
def _get_client():
|
|
|
|
| 34 |
AUTOCOMPLETE_REPO = os.environ.get("AUTOCOMPLETE_REPO_ID", "bayan10/AutoComplete")
|
| 35 |
|
| 36 |
|
| 37 |
+
def _find_working_endpoint():
|
| 38 |
+
"""Try multiple HF API endpoints to find one that resolves."""
|
| 39 |
+
global _working_url
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
+
if _working_url:
|
| 42 |
+
return _working_url
|
| 43 |
+
|
| 44 |
+
# Candidate API endpoints
|
| 45 |
+
candidates = [
|
| 46 |
+
"https://router.huggingface.co/hf-inference/models/",
|
| 47 |
+
"https://api-inference.huggingface.co/models/",
|
| 48 |
+
"https://api.huggingface.co/models/",
|
| 49 |
+
"https://huggingface.co/api/models/",
|
| 50 |
+
]
|
| 51 |
+
|
| 52 |
+
for url in candidates:
|
| 53 |
+
# Extract hostname from URL
|
| 54 |
+
hostname = url.split("//")[1].split("/")[0]
|
| 55 |
+
try:
|
| 56 |
+
socket.getaddrinfo(hostname, 443)
|
| 57 |
+
logger.info("DNS resolved for: %s", hostname)
|
| 58 |
+
_working_url = url
|
| 59 |
+
return url
|
| 60 |
+
except socket.gaierror:
|
| 61 |
+
logger.warning("DNS failed for: %s", hostname)
|
| 62 |
+
continue
|
| 63 |
+
|
| 64 |
+
logger.error("No HF API endpoint is reachable!")
|
| 65 |
+
return None
|
| 66 |
+
|
| 67 |
+
|
| 68 |
+
def _call_model_httpx(repo_id, payload, task=""):
|
| 69 |
+
"""Call HF model using httpx (same transport as InferenceClient)."""
|
| 70 |
+
import httpx
|
| 71 |
+
|
| 72 |
+
base_url = _find_working_endpoint()
|
| 73 |
+
if not base_url:
|
| 74 |
+
raise RuntimeError("No reachable HF API endpoint found")
|
| 75 |
+
|
| 76 |
+
url = base_url + repo_id
|
| 77 |
|
| 78 |
if "options" not in payload:
|
| 79 |
payload["options"] = {"wait_for_model": True}
|
| 80 |
|
| 81 |
+
headers = {"Content-Type": "application/json"}
|
| 82 |
+
if HF_API_TOKEN:
|
| 83 |
+
headers["Authorization"] = "Bearer " + HF_API_TOKEN
|
| 84 |
|
| 85 |
+
logger.info("Calling HF model: %s at %s", repo_id, url)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
|
| 87 |
+
with httpx.Client(timeout=120) as client:
|
| 88 |
+
resp = client.post(url, json=payload, headers=headers)
|
| 89 |
|
| 90 |
+
if resp.status_code != 200:
|
| 91 |
+
raise RuntimeError(f"HF API error (HTTP {resp.status_code}): {resp.text[:300]}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 92 |
|
| 93 |
+
result = resp.json()
|
| 94 |
logger.info("HF result for %s: type=%s preview=%s",
|
| 95 |
repo_id, type(result).__name__, str(result)[:200])
|
| 96 |
|
|
|
|
| 97 |
if isinstance(result, dict) and "error" in result:
|
| 98 |
raise RuntimeError("HF API error: " + str(result["error"]))
|
| 99 |
|
|
|
|
| 123 |
# ============================================================
|
| 124 |
|
| 125 |
def hf_summarize(text, max_length=128, min_length=30):
|
| 126 |
+
result = _call_model_httpx(SUMMARIZATION_REPO, {
|
| 127 |
"inputs": text,
|
| 128 |
"parameters": {"max_length": max_length, "min_length": min_length},
|
| 129 |
}, task="summarization")
|
|
|
|
| 131 |
|
| 132 |
|
| 133 |
def hf_correct_spelling(text):
|
| 134 |
+
result = _call_model_httpx(SPELLING_REPO, {"inputs": text})
|
| 135 |
return _extract_text(result, text)
|
| 136 |
|
| 137 |
|
| 138 |
def hf_add_punctuation(text):
|
| 139 |
+
result = _call_model_httpx(PUNCTUATION_REPO, {"inputs": text})
|
| 140 |
return _extract_text(result, text)
|
| 141 |
|
| 142 |
|
| 143 |
def hf_autocomplete(text, n=5):
|
| 144 |
+
result = _call_model_httpx(AUTOCOMPLETE_REPO, {
|
| 145 |
"inputs": text,
|
| 146 |
"parameters": {"max_new_tokens": 20},
|
| 147 |
}, task="text-generation")
|
|
|
|
| 163 |
|
| 164 |
def check_hf_api_available():
|
| 165 |
try:
|
| 166 |
+
return _find_working_endpoint() is not None
|
| 167 |
except Exception:
|
| 168 |
return False
|
| 169 |
|
| 170 |
|
| 171 |
def debug_test_all_models():
|
| 172 |
+
"""Test DNS resolution + all HF models."""
|
| 173 |
results = {}
|
| 174 |
test_text = "هذا نص تجريبي للاختبار"
|
| 175 |
long_text = (test_text + " ") * 5
|
| 176 |
|
| 177 |
+
# DNS diagnostics
|
| 178 |
+
dns_results = {}
|
| 179 |
+
hostnames = [
|
| 180 |
+
"router.huggingface.co",
|
| 181 |
+
"api-inference.huggingface.co",
|
| 182 |
+
"api.huggingface.co",
|
| 183 |
+
"huggingface.co",
|
| 184 |
+
"hf.co",
|
| 185 |
+
"google.com",
|
| 186 |
+
"pypi.org",
|
| 187 |
+
]
|
| 188 |
+
for hostname in hostnames:
|
| 189 |
+
try:
|
| 190 |
+
addrs = socket.getaddrinfo(hostname, 443)
|
| 191 |
+
dns_results[hostname] = "OK (" + addrs[0][4][0] + ")"
|
| 192 |
+
except socket.gaierror as e:
|
| 193 |
+
dns_results[hostname] = "FAIL: " + str(e)
|
| 194 |
+
|
| 195 |
+
results["_dns"] = dns_results
|
| 196 |
+
results["_working_endpoint"] = _find_working_endpoint() or "NONE"
|
| 197 |
+
|
| 198 |
try:
|
| 199 |
import huggingface_hub
|
| 200 |
results["_info"] = {"hf_hub_version": huggingface_hub.__version__}
|