Spaces:
Running
Running
add cv processing fallback
Browse files- config.py +1 -0
- requirements.txt +1 -0
- services/new/cv_markdown_processor.py +1 -1
- services/new/cv_processor_v2.py +1 -1
- services/new/google/gemini_extractor.py +93 -44
config.py
CHANGED
|
@@ -16,6 +16,7 @@ load_dotenv(dotenv_path=env_path)
|
|
| 16 |
# Google Gemini
|
| 17 |
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY", "")
|
| 18 |
GEMINI_MODEL = "gemini-3.5-flash-lite"
|
|
|
|
| 19 |
|
| 20 |
MAX_OUTPUT_TOKENS = 4096
|
| 21 |
TEMPERATURE = 0.1
|
|
|
|
| 16 |
# Google Gemini
|
| 17 |
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY", "")
|
| 18 |
GEMINI_MODEL = "gemini-3.5-flash-lite"
|
| 19 |
+
GEMINI_FALLBACK_MODEL = "gemini-3.1-flash-lite"
|
| 20 |
|
| 21 |
MAX_OUTPUT_TOKENS = 4096
|
| 22 |
TEMPERATURE = 0.1
|
requirements.txt
CHANGED
|
@@ -17,6 +17,7 @@ pypandoc
|
|
| 17 |
pandoc
|
| 18 |
|
| 19 |
# V2 — lightweight PDF extraction (no ML models)
|
|
|
|
| 20 |
pymupdf4llm
|
| 21 |
|
| 22 |
# V2 — local LLM inference + model download
|
|
|
|
| 17 |
pandoc
|
| 18 |
|
| 19 |
# V2 — lightweight PDF extraction (no ML models)
|
| 20 |
+
pymupdf
|
| 21 |
pymupdf4llm
|
| 22 |
|
| 23 |
# V2 — local LLM inference + model download
|
services/new/cv_markdown_processor.py
CHANGED
|
@@ -40,7 +40,7 @@ class CvMarkdownProcessor:
|
|
| 40 |
return finalize_v2_cv_payload(
|
| 41 |
markdown=markdown,
|
| 42 |
extracted=extracted,
|
| 43 |
-
model_used=GEMINI_MODEL,
|
| 44 |
file_type="markdown",
|
| 45 |
is_scanned=False,
|
| 46 |
page_count=1,
|
|
|
|
| 40 |
return finalize_v2_cv_payload(
|
| 41 |
markdown=markdown,
|
| 42 |
extracted=extracted,
|
| 43 |
+
model_used=extracted.get("model_used", GEMINI_MODEL),
|
| 44 |
file_type="markdown",
|
| 45 |
is_scanned=False,
|
| 46 |
page_count=1,
|
services/new/cv_processor_v2.py
CHANGED
|
@@ -69,7 +69,7 @@ class CvProcessorV2:
|
|
| 69 |
return finalize_v2_cv_payload(
|
| 70 |
markdown=markdown,
|
| 71 |
extracted=extracted,
|
| 72 |
-
model_used=GEMINI_MODEL,
|
| 73 |
file_type=conversion.file_type,
|
| 74 |
is_scanned=conversion.is_scanned,
|
| 75 |
page_count=conversion.page_count,
|
|
|
|
| 69 |
return finalize_v2_cv_payload(
|
| 70 |
markdown=markdown,
|
| 71 |
extracted=extracted,
|
| 72 |
+
model_used=extracted.get("model_used", GEMINI_MODEL),
|
| 73 |
file_type=conversion.file_type,
|
| 74 |
is_scanned=conversion.is_scanned,
|
| 75 |
page_count=conversion.page_count,
|
services/new/google/gemini_extractor.py
CHANGED
|
@@ -16,6 +16,7 @@ import httpx
|
|
| 16 |
|
| 17 |
from config import (
|
| 18 |
GEMINI_API_KEY,
|
|
|
|
| 19 |
GEMINI_MODEL,
|
| 20 |
MAX_OUTPUT_TOKENS,
|
| 21 |
TEMPERATURE,
|
|
@@ -119,14 +120,80 @@ CV MARKDOWN:
|
|
| 119 |
|
| 120 |
|
| 121 |
class GeminiCvExtractor:
|
| 122 |
-
"""
|
| 123 |
-
|
| 124 |
-
def __init__(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 125 |
self.api_key = api_key or GEMINI_API_KEY
|
| 126 |
self.model = model or GEMINI_MODEL
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 127 |
|
| 128 |
def _call_interactions(self, prompt: str, system_instruction: str = "") -> dict:
|
| 129 |
-
"""Call /v1beta/interactions with
|
| 130 |
if not self.api_key or self.api_key == "YOUR_GEMINI_API_KEY_HERE":
|
| 131 |
raise ValueError("GEMINI_API_KEY is not set in processor/.env")
|
| 132 |
|
|
@@ -135,53 +202,34 @@ class GeminiCvExtractor:
|
|
| 135 |
"x-goog-api-key": self.api_key,
|
| 136 |
"Content-Type": "application/json",
|
| 137 |
}
|
| 138 |
-
body: Dict[str, Any] = {
|
| 139 |
-
"model": self.model,
|
| 140 |
-
"input": prompt,
|
| 141 |
-
}
|
| 142 |
-
if system_instruction:
|
| 143 |
-
body["system_instruction"] = system_instruction
|
| 144 |
|
| 145 |
-
|
| 146 |
-
|
|
|
|
| 147 |
|
|
|
|
| 148 |
with httpx.Client(timeout=60.0) as client:
|
| 149 |
-
for
|
| 150 |
try:
|
| 151 |
-
|
| 152 |
-
|
| 153 |
-
|
| 154 |
-
|
| 155 |
-
|
| 156 |
-
|
| 157 |
-
|
| 158 |
-
|
| 159 |
-
|
| 160 |
-
|
| 161 |
-
|
| 162 |
-
"usage": data.get("usage", {}),
|
| 163 |
-
}
|
| 164 |
-
|
| 165 |
-
last_error = f"Gemini Interactions API error ({resp.status_code}): {resp.text}"
|
| 166 |
-
if resp.status_code in (429, 500, 502, 503, 504) and attempt < max_retries:
|
| 167 |
-
sleep_time = attempt * 1.5
|
| 168 |
logger.warning(
|
| 169 |
-
"Gemini
|
| 170 |
-
|
|
|
|
|
|
|
| 171 |
)
|
| 172 |
-
time.sleep(sleep_time)
|
| 173 |
-
continue
|
| 174 |
-
else:
|
| 175 |
-
raise RuntimeError(last_error)
|
| 176 |
-
|
| 177 |
-
except httpx.RequestError as exc:
|
| 178 |
-
last_error = f"Network error during Gemini call: {exc}"
|
| 179 |
-
if attempt < max_retries:
|
| 180 |
-
time.sleep(attempt * 1.5)
|
| 181 |
-
continue
|
| 182 |
-
raise RuntimeError(last_error)
|
| 183 |
|
| 184 |
-
|
| 185 |
|
| 186 |
def ping(self, prompt: str = "Say hello in one short sentence.") -> Dict[str, Any]:
|
| 187 |
"""Test API key and latency directly."""
|
|
@@ -205,6 +253,7 @@ class GeminiCvExtractor:
|
|
| 205 |
data["completeness_score"] = self._completeness(data)
|
| 206 |
data["markdown"] = markdown
|
| 207 |
data["usage_metadata"] = res.get("usage", {})
|
|
|
|
| 208 |
return data
|
| 209 |
|
| 210 |
@staticmethod
|
|
|
|
| 16 |
|
| 17 |
from config import (
|
| 18 |
GEMINI_API_KEY,
|
| 19 |
+
GEMINI_FALLBACK_MODEL,
|
| 20 |
GEMINI_MODEL,
|
| 21 |
MAX_OUTPUT_TOKENS,
|
| 22 |
TEMPERATURE,
|
|
|
|
| 120 |
|
| 121 |
|
| 122 |
class GeminiCvExtractor:
|
| 123 |
+
"""Calls Google Gemini Interactions API with model-level fallback."""
|
| 124 |
+
|
| 125 |
+
def __init__(
|
| 126 |
+
self,
|
| 127 |
+
api_key: Optional[str] = None,
|
| 128 |
+
model: Optional[str] = None,
|
| 129 |
+
fallback_model: Optional[str] = None,
|
| 130 |
+
):
|
| 131 |
self.api_key = api_key or GEMINI_API_KEY
|
| 132 |
self.model = model or GEMINI_MODEL
|
| 133 |
+
self.fallback_model = (
|
| 134 |
+
fallback_model if fallback_model is not None else GEMINI_FALLBACK_MODEL
|
| 135 |
+
)
|
| 136 |
+
|
| 137 |
+
def _call_model(
|
| 138 |
+
self,
|
| 139 |
+
client: httpx.Client,
|
| 140 |
+
url: str,
|
| 141 |
+
headers: Dict[str, str],
|
| 142 |
+
model_name: str,
|
| 143 |
+
prompt: str,
|
| 144 |
+
system_instruction: str = "",
|
| 145 |
+
) -> dict:
|
| 146 |
+
"""Call /v1beta/interactions for a single model with up to 2 attempts."""
|
| 147 |
+
body: Dict[str, Any] = {
|
| 148 |
+
"model": model_name,
|
| 149 |
+
"input": prompt,
|
| 150 |
+
}
|
| 151 |
+
if system_instruction:
|
| 152 |
+
body["system_instruction"] = system_instruction
|
| 153 |
+
|
| 154 |
+
max_retries = 2
|
| 155 |
+
last_error = ""
|
| 156 |
+
|
| 157 |
+
for attempt in range(1, max_retries + 1):
|
| 158 |
+
try:
|
| 159 |
+
resp = client.post(url, headers=headers, json=body)
|
| 160 |
+
if resp.status_code == 200:
|
| 161 |
+
data = resp.json()
|
| 162 |
+
reply = ""
|
| 163 |
+
for step in data.get("steps", []):
|
| 164 |
+
for item in step.get("content", []):
|
| 165 |
+
if item.get("type") == "text":
|
| 166 |
+
reply += item.get("text", "")
|
| 167 |
+
|
| 168 |
+
return {
|
| 169 |
+
"text": reply.strip(),
|
| 170 |
+
"usage": data.get("usage", {}),
|
| 171 |
+
"model": model_name,
|
| 172 |
+
}
|
| 173 |
+
|
| 174 |
+
last_error = f"Gemini Interactions API error ({resp.status_code}): {resp.text}"
|
| 175 |
+
if resp.status_code in (429, 500, 502, 503, 504) and attempt < max_retries:
|
| 176 |
+
sleep_time = attempt * 1.5
|
| 177 |
+
logger.warning(
|
| 178 |
+
"Gemini (%s) attempt %d/%d failed (%d). Retrying in %.1fs...",
|
| 179 |
+
model_name, attempt, max_retries, resp.status_code, sleep_time
|
| 180 |
+
)
|
| 181 |
+
time.sleep(sleep_time)
|
| 182 |
+
continue
|
| 183 |
+
else:
|
| 184 |
+
raise RuntimeError(last_error)
|
| 185 |
+
|
| 186 |
+
except httpx.RequestError as exc:
|
| 187 |
+
last_error = f"Network error during Gemini call: {exc}"
|
| 188 |
+
if attempt < max_retries:
|
| 189 |
+
time.sleep(attempt * 1.5)
|
| 190 |
+
continue
|
| 191 |
+
raise RuntimeError(last_error)
|
| 192 |
+
|
| 193 |
+
raise RuntimeError(last_error or f"Gemini API failed after {max_retries} attempts.")
|
| 194 |
|
| 195 |
def _call_interactions(self, prompt: str, system_instruction: str = "") -> dict:
|
| 196 |
+
"""Call /v1beta/interactions with 2 attempts per model and fallback to GEMINI_FALLBACK_MODEL."""
|
| 197 |
if not self.api_key or self.api_key == "YOUR_GEMINI_API_KEY_HERE":
|
| 198 |
raise ValueError("GEMINI_API_KEY is not set in processor/.env")
|
| 199 |
|
|
|
|
| 202 |
"x-goog-api-key": self.api_key,
|
| 203 |
"Content-Type": "application/json",
|
| 204 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 205 |
|
| 206 |
+
models = [self.model]
|
| 207 |
+
if self.fallback_model and self.fallback_model != self.model:
|
| 208 |
+
models.append(self.fallback_model)
|
| 209 |
|
| 210 |
+
last_exc: Optional[Exception] = None
|
| 211 |
with httpx.Client(timeout=60.0) as client:
|
| 212 |
+
for idx, model_name in enumerate(models):
|
| 213 |
try:
|
| 214 |
+
return self._call_model(
|
| 215 |
+
client=client,
|
| 216 |
+
url=url,
|
| 217 |
+
headers=headers,
|
| 218 |
+
model_name=model_name,
|
| 219 |
+
prompt=prompt,
|
| 220 |
+
system_instruction=system_instruction,
|
| 221 |
+
)
|
| 222 |
+
except Exception as exc:
|
| 223 |
+
last_exc = exc
|
| 224 |
+
if idx + 1 < len(models):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 225 |
logger.warning(
|
| 226 |
+
"Gemini model '%s' failed (%s). Falling back to '%s'...",
|
| 227 |
+
model_name,
|
| 228 |
+
exc,
|
| 229 |
+
models[idx + 1],
|
| 230 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 231 |
|
| 232 |
+
raise last_exc or RuntimeError("Gemini API call failed.")
|
| 233 |
|
| 234 |
def ping(self, prompt: str = "Say hello in one short sentence.") -> Dict[str, Any]:
|
| 235 |
"""Test API key and latency directly."""
|
|
|
|
| 253 |
data["completeness_score"] = self._completeness(data)
|
| 254 |
data["markdown"] = markdown
|
| 255 |
data["usage_metadata"] = res.get("usage", {})
|
| 256 |
+
data["model_used"] = res.get("model", self.model)
|
| 257 |
return data
|
| 258 |
|
| 259 |
@staticmethod
|