Spaces:
Sleeping
Sleeping
Añade funciones para manejar imágenes y normalizar contenido de usuario en `app.py`. Se implementan las funciones `_build_image_parts` y `_value_to_user_content` para procesar imágenes y generar contenido de usuario adecuado. Además, se mejora la función `_value_preview` para ofrecer una vista previa segura de los mensajes, incluyendo el conteo de imágenes. Se actualiza la función `respond` para utilizar estas nuevas funciones, optimizando el manejo de contenido y mejorando la legibilidad del código.
Browse files
app.py
CHANGED
|
@@ -92,6 +92,53 @@ def _extract_text_and_files(message):
|
|
| 92 |
return text_combined, files
|
| 93 |
|
| 94 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
def _preview_text(text: str | None, limit: int = 600) -> str:
|
| 96 |
if not text:
|
| 97 |
return ""
|
|
@@ -107,9 +154,9 @@ def _history_preview(history: list[tuple[str, str]] | None, max_turns: int = 3,
|
|
| 107 |
parts: list[str] = []
|
| 108 |
for user_turn, assistant_turn in tail:
|
| 109 |
if user_turn:
|
| 110 |
-
parts.append(f"
|
| 111 |
if assistant_turn:
|
| 112 |
-
parts.append(f"
|
| 113 |
joined = "\n".join(parts)
|
| 114 |
return _preview_text(joined, max_chars)
|
| 115 |
|
|
@@ -138,36 +185,15 @@ def respond(message, history: list[tuple[str, str]]):
|
|
| 138 |
]
|
| 139 |
for user_turn, assistant_turn in history or []:
|
| 140 |
if user_turn:
|
| 141 |
-
messages.append({"role": "user", "content": user_turn})
|
| 142 |
if assistant_turn:
|
| 143 |
messages.append({"role": "assistant", "content": assistant_turn})
|
| 144 |
|
| 145 |
# Build user content with optional inline images (data URLs)
|
| 146 |
final_user_text = (user_text or "").strip() or "Describe el contenido de la(s) imagen(es)."
|
| 147 |
|
| 148 |
-
# Collect image parts
|
| 149 |
-
image_parts =
|
| 150 |
-
for f in files or []:
|
| 151 |
-
path = None
|
| 152 |
-
if isinstance(f, str):
|
| 153 |
-
path = f
|
| 154 |
-
elif isinstance(f, dict):
|
| 155 |
-
path = f.get("path") or f.get("name")
|
| 156 |
-
if not path or not os.path.exists(path):
|
| 157 |
-
continue
|
| 158 |
-
mime, _ = mimetypes.guess_type(path)
|
| 159 |
-
if not mime or not mime.startswith("image/"):
|
| 160 |
-
continue
|
| 161 |
-
try:
|
| 162 |
-
with open(path, "rb") as fp:
|
| 163 |
-
b64 = base64.b64encode(fp.read()).decode("utf-8")
|
| 164 |
-
data_url = f"data:{mime};base64,{b64}"
|
| 165 |
-
image_parts.append({
|
| 166 |
-
"type": "image_url",
|
| 167 |
-
"image_url": {"url": data_url},
|
| 168 |
-
})
|
| 169 |
-
except Exception:
|
| 170 |
-
continue
|
| 171 |
|
| 172 |
if image_parts:
|
| 173 |
user_content = [{"type": "text", "text": final_user_text}] + image_parts
|
|
@@ -187,7 +213,7 @@ def respond(message, history: list[tuple[str, str]]):
|
|
| 187 |
name="Chat Session",
|
| 188 |
run_type="chain",
|
| 189 |
inputs={
|
| 190 |
-
"user_text":
|
| 191 |
"has_images": bool(image_parts),
|
| 192 |
"history_preview": _history_preview(history),
|
| 193 |
},
|
|
|
|
| 92 |
return text_combined, files
|
| 93 |
|
| 94 |
|
| 95 |
+
def _build_image_parts(files):
|
| 96 |
+
image_parts = []
|
| 97 |
+
for f in files or []:
|
| 98 |
+
path = None
|
| 99 |
+
if isinstance(f, str):
|
| 100 |
+
path = f
|
| 101 |
+
elif isinstance(f, dict):
|
| 102 |
+
path = f.get("path") or f.get("name")
|
| 103 |
+
if not path or not os.path.exists(path):
|
| 104 |
+
continue
|
| 105 |
+
mime, _ = mimetypes.guess_type(path)
|
| 106 |
+
if not mime or not mime.startswith("image/"):
|
| 107 |
+
continue
|
| 108 |
+
try:
|
| 109 |
+
with open(path, "rb") as fp:
|
| 110 |
+
b64 = base64.b64encode(fp.read()).decode("utf-8")
|
| 111 |
+
data_url = f"data:{mime};base64,{b64}"
|
| 112 |
+
image_parts.append({
|
| 113 |
+
"type": "image_url",
|
| 114 |
+
"image_url": {"url": data_url},
|
| 115 |
+
})
|
| 116 |
+
except Exception:
|
| 117 |
+
continue
|
| 118 |
+
return image_parts
|
| 119 |
+
|
| 120 |
+
|
| 121 |
+
def _value_to_user_content(value):
|
| 122 |
+
"""Normalize any gradio message value to OpenAI user 'content'."""
|
| 123 |
+
text, files = _extract_text_and_files(value)
|
| 124 |
+
final_user_text = (text or "").strip() or "Describe el contenido de la(s) imagen(es)."
|
| 125 |
+
image_parts = _build_image_parts(files)
|
| 126 |
+
if image_parts:
|
| 127 |
+
return [{"type": "text", "text": final_user_text}] + image_parts
|
| 128 |
+
return final_user_text
|
| 129 |
+
|
| 130 |
+
|
| 131 |
+
def _value_preview(value, limit: int = 600) -> str:
|
| 132 |
+
"""Safe preview string for any kind of message value."""
|
| 133 |
+
if isinstance(value, str):
|
| 134 |
+
return _preview_text(value, limit)
|
| 135 |
+
text, files = _extract_text_and_files(value)
|
| 136 |
+
suffix = ""
|
| 137 |
+
if files:
|
| 138 |
+
suffix = f" [images:{len(files)}]"
|
| 139 |
+
return _preview_text((text or "").strip() + suffix, limit)
|
| 140 |
+
|
| 141 |
+
|
| 142 |
def _preview_text(text: str | None, limit: int = 600) -> str:
|
| 143 |
if not text:
|
| 144 |
return ""
|
|
|
|
| 154 |
parts: list[str] = []
|
| 155 |
for user_turn, assistant_turn in tail:
|
| 156 |
if user_turn:
|
| 157 |
+
parts.append(f"User 👤: {_preview_text(user_turn, 300)}")
|
| 158 |
if assistant_turn:
|
| 159 |
+
parts.append(f"Assistant 🤖: {_preview_text(assistant_turn, 300)}")
|
| 160 |
joined = "\n".join(parts)
|
| 161 |
return _preview_text(joined, max_chars)
|
| 162 |
|
|
|
|
| 185 |
]
|
| 186 |
for user_turn, assistant_turn in history or []:
|
| 187 |
if user_turn:
|
| 188 |
+
messages.append({"role": "user", "content": _value_to_user_content(user_turn)})
|
| 189 |
if assistant_turn:
|
| 190 |
messages.append({"role": "assistant", "content": assistant_turn})
|
| 191 |
|
| 192 |
# Build user content with optional inline images (data URLs)
|
| 193 |
final_user_text = (user_text or "").strip() or "Describe el contenido de la(s) imagen(es)."
|
| 194 |
|
| 195 |
+
# Collect image parts using helper
|
| 196 |
+
image_parts = _build_image_parts(files)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 197 |
|
| 198 |
if image_parts:
|
| 199 |
user_content = [{"type": "text", "text": final_user_text}] + image_parts
|
|
|
|
| 213 |
name="Chat Session",
|
| 214 |
run_type="chain",
|
| 215 |
inputs={
|
| 216 |
+
"user_text": _value_preview(message, 600),
|
| 217 |
"has_images": bool(image_parts),
|
| 218 |
"history_preview": _history_preview(history),
|
| 219 |
},
|