Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -24,6 +24,7 @@ except Exception:
|
|
24 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
25 |
from huggingface_hub import login, HfApi
|
26 |
|
|
|
27 |
# -------------------
|
28 |
# Configuration
|
29 |
# -------------------
|
@@ -37,8 +38,9 @@ HF_TOKEN = (
|
|
37 |
COHERE_API_KEY = os.getenv("COHERE_API_KEY")
|
38 |
USE_HOSTED_COHERE = bool(COHERE_API_KEY and _HAS_COHERE)
|
39 |
|
|
|
40 |
# -------------------
|
41 |
-
# Helpers (for
|
42 |
# -------------------
|
43 |
def local_now_str(user_tz: str | None) -> tuple[str, str]:
|
44 |
"""Returns (label, formatted_time). Falls back to UTC if tz missing/invalid."""
|
@@ -54,7 +56,9 @@ def local_now_str(user_tz: str | None) -> tuple[str, str]:
|
|
54 |
label = "UTC"
|
55 |
return label, dt.strftime("%Y-%m-%d %H:%M:%S")
|
56 |
|
|
|
57 |
def header(processing_time=None, user_tz: str | None = None):
|
|
|
58 |
tz_label, now_str = local_now_str(user_tz)
|
59 |
s = (
|
60 |
f"Current Date and Time ({tz_label} - YYYY-MM-DD HH:MM:SS formatted): {now_str}\n"
|
@@ -64,6 +68,7 @@ def header(processing_time=None, user_tz: str | None = None):
|
|
64 |
s += f"Processing Time: {processing_time:.2f} seconds\n"
|
65 |
return s
|
66 |
|
|
|
67 |
def pick_dtype_and_map():
|
68 |
if torch.cuda.is_available():
|
69 |
return torch.float16, "auto"
|
@@ -71,6 +76,7 @@ def pick_dtype_and_map():
|
|
71 |
return torch.float16, {"": "mps"}
|
72 |
return torch.float32, "cpu" # CPU path (likely too big for R7B)
|
73 |
|
|
|
74 |
def is_identity_query(message: str, history) -> bool:
|
75 |
"""Detects identity questions in current message or most recent user turn."""
|
76 |
patterns = [
|
@@ -85,17 +91,23 @@ def is_identity_query(message: str, history) -> bool:
|
|
85 |
r"\byour\s+name\b",
|
86 |
r"\bwho\s+am\s+i\s+chatting\s+with\b",
|
87 |
]
|
|
|
88 |
def hit(text: str | None) -> bool:
|
89 |
t = (text or "").strip().lower()
|
90 |
return any(re.search(p, t) for p in patterns)
|
|
|
91 |
if hit(message):
|
92 |
return True
|
|
|
93 |
if history:
|
|
|
94 |
last_user = history[-1][0] if isinstance(history[-1], (list, tuple)) and history[-1] else None
|
95 |
if hit(last_user):
|
96 |
return True
|
|
|
97 |
return False
|
98 |
|
|
|
99 |
# -------------------
|
100 |
# Cohere Hosted Path
|
101 |
# -------------------
|
@@ -103,6 +115,7 @@ _co_client = None
|
|
103 |
if USE_HOSTED_COHERE:
|
104 |
_co_client = cohere.Client(api_key=COHERE_API_KEY)
|
105 |
|
|
|
106 |
def _cohere_parse(resp):
|
107 |
# v5+ responses.create
|
108 |
if hasattr(resp, "output_text") and resp.output_text:
|
@@ -116,6 +129,7 @@ def _cohere_parse(resp):
|
|
116 |
return resp.text.strip()
|
117 |
return "Sorry, I couldn't parse the response from Cohere."
|
118 |
|
|
|
119 |
def cohere_chat(message, history):
|
120 |
try:
|
121 |
# Prefer modern API
|
@@ -143,6 +157,7 @@ def cohere_chat(message, history):
|
|
143 |
except Exception as e:
|
144 |
return f"Error calling Cohere API: {e}"
|
145 |
|
|
|
146 |
# -------------------
|
147 |
# Local HF Path
|
148 |
# -------------------
|
@@ -153,20 +168,31 @@ def load_local_model():
|
|
153 |
"HUGGINGFACE_HUB_TOKEN (or HF_TOKEN) is not set. "
|
154 |
"Either set it, or provide COHERE_API_KEY to use Cohere's hosted API."
|
155 |
)
|
|
|
156 |
login(token=HF_TOKEN, add_to_git_credential=False)
|
|
|
157 |
dtype, device_map = pick_dtype_and_map()
|
158 |
tok = AutoTokenizer.from_pretrained(
|
159 |
-
MODEL_ID,
|
160 |
-
|
|
|
|
|
|
|
|
|
161 |
)
|
162 |
mdl = AutoModelForCausalLM.from_pretrained(
|
163 |
-
MODEL_ID,
|
164 |
-
|
|
|
|
|
|
|
|
|
165 |
)
|
166 |
if mdl.config.eos_token_id is None and tok.eos_token_id is not None:
|
167 |
mdl.config.eos_token_id = tok.eos_token_id
|
168 |
return mdl, tok
|
169 |
|
|
|
170 |
def build_inputs(tokenizer, message, history):
|
171 |
msgs = []
|
172 |
for u, a in (history or []):
|
@@ -177,6 +203,7 @@ def build_inputs(tokenizer, message, history):
|
|
177 |
msgs, tokenize=True, add_generation_prompt=True, return_tensors="pt"
|
178 |
)
|
179 |
|
|
|
180 |
def local_generate(model, tokenizer, input_ids, max_new_tokens=350):
|
181 |
input_ids = input_ids.to(model.device)
|
182 |
with torch.no_grad():
|
@@ -194,6 +221,7 @@ def local_generate(model, tokenizer, input_ids, max_new_tokens=350):
|
|
194 |
text = tokenizer.decode(gen_only, skip_special_tokens=True)
|
195 |
return text.strip()
|
196 |
|
|
|
197 |
# -------------------
|
198 |
# Chat callback (no header/meta in chat replies)
|
199 |
# -------------------
|
@@ -218,6 +246,7 @@ def chat_fn(message, history, user_tz):
|
|
218 |
except Exception as e:
|
219 |
return f"Error during chat: {e}"
|
220 |
|
|
|
221 |
# -------------------
|
222 |
# Connection check (keeps header/meta)
|
223 |
# -------------------
|
@@ -243,22 +272,23 @@ def check_connection(user_tz=None):
|
|
243 |
except Exception as e:
|
244 |
return f"{header(user_tz=user_tz)}Connection Status: ❌ Error\nDetails: {e}"
|
245 |
|
|
|
246 |
# -------------------
|
247 |
# UI
|
248 |
# -------------------
|
249 |
with gr.Blocks(theme=gr.themes.Default()) as demo:
|
250 |
-
#
|
251 |
user_tz_state = gr.State("")
|
252 |
-
# On load, capture browser timezone via JS and store in user_tz_state
|
253 |
-
demo.load(
|
254 |
-
fn=lambda tz: tz, # echo the JS value back to Gradio
|
255 |
-
inputs=None,
|
256 |
-
outputs=[user_tz_state], # outputs must be a LIST
|
257 |
-
js="() => Intl.DateTimeFormat().resolvedOptions().timeZone"
|
258 |
-
)
|
259 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
260 |
|
261 |
-
gr.Markdown(
|
262 |
|
263 |
with gr.Row():
|
264 |
btn = gr.Button("Check Connection Status")
|
@@ -273,7 +303,7 @@ demo.load(
|
|
273 |
chat = gr.ChatInterface(
|
274 |
fn=chat_fn,
|
275 |
type="messages",
|
276 |
-
additional_inputs=[user_tz_state], # pass timezone into chat_fn
|
277 |
description="A medical decision support system that provides healthcare-related information and decision making support.",
|
278 |
examples=[
|
279 |
["What are the symptoms of hypertension?", ""],
|
@@ -283,12 +313,10 @@ demo.load(
|
|
283 |
cache_examples=False,
|
284 |
)
|
285 |
|
|
|
286 |
btn.click(fn=check_connection, inputs=user_tz_state, outputs=status)
|
287 |
|
288 |
if __name__ == "__main__":
|
289 |
demo.launch()
|
290 |
|
291 |
|
292 |
-
|
293 |
-
|
294 |
-
|
|
|
24 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
25 |
from huggingface_hub import login, HfApi
|
26 |
|
27 |
+
|
28 |
# -------------------
|
29 |
# Configuration
|
30 |
# -------------------
|
|
|
38 |
COHERE_API_KEY = os.getenv("COHERE_API_KEY")
|
39 |
USE_HOSTED_COHERE = bool(COHERE_API_KEY and _HAS_COHERE)
|
40 |
|
41 |
+
|
42 |
# -------------------
|
43 |
+
# Helpers (used for the connection card only)
|
44 |
# -------------------
|
45 |
def local_now_str(user_tz: str | None) -> tuple[str, str]:
|
46 |
"""Returns (label, formatted_time). Falls back to UTC if tz missing/invalid."""
|
|
|
56 |
label = "UTC"
|
57 |
return label, dt.strftime("%Y-%m-%d %H:%M:%S")
|
58 |
|
59 |
+
|
60 |
def header(processing_time=None, user_tz: str | None = None):
|
61 |
+
"""Only used in the connection status panel (not in chat replies)."""
|
62 |
tz_label, now_str = local_now_str(user_tz)
|
63 |
s = (
|
64 |
f"Current Date and Time ({tz_label} - YYYY-MM-DD HH:MM:SS formatted): {now_str}\n"
|
|
|
68 |
s += f"Processing Time: {processing_time:.2f} seconds\n"
|
69 |
return s
|
70 |
|
71 |
+
|
72 |
def pick_dtype_and_map():
|
73 |
if torch.cuda.is_available():
|
74 |
return torch.float16, "auto"
|
|
|
76 |
return torch.float16, {"": "mps"}
|
77 |
return torch.float32, "cpu" # CPU path (likely too big for R7B)
|
78 |
|
79 |
+
|
80 |
def is_identity_query(message: str, history) -> bool:
|
81 |
"""Detects identity questions in current message or most recent user turn."""
|
82 |
patterns = [
|
|
|
91 |
r"\byour\s+name\b",
|
92 |
r"\bwho\s+am\s+i\s+chatting\s+with\b",
|
93 |
]
|
94 |
+
|
95 |
def hit(text: str | None) -> bool:
|
96 |
t = (text or "").strip().lower()
|
97 |
return any(re.search(p, t) for p in patterns)
|
98 |
+
|
99 |
if hit(message):
|
100 |
return True
|
101 |
+
|
102 |
if history:
|
103 |
+
# Gradio history: List[Tuple[user, assistant]]
|
104 |
last_user = history[-1][0] if isinstance(history[-1], (list, tuple)) and history[-1] else None
|
105 |
if hit(last_user):
|
106 |
return True
|
107 |
+
|
108 |
return False
|
109 |
|
110 |
+
|
111 |
# -------------------
|
112 |
# Cohere Hosted Path
|
113 |
# -------------------
|
|
|
115 |
if USE_HOSTED_COHERE:
|
116 |
_co_client = cohere.Client(api_key=COHERE_API_KEY)
|
117 |
|
118 |
+
|
119 |
def _cohere_parse(resp):
|
120 |
# v5+ responses.create
|
121 |
if hasattr(resp, "output_text") and resp.output_text:
|
|
|
129 |
return resp.text.strip()
|
130 |
return "Sorry, I couldn't parse the response from Cohere."
|
131 |
|
132 |
+
|
133 |
def cohere_chat(message, history):
|
134 |
try:
|
135 |
# Prefer modern API
|
|
|
157 |
except Exception as e:
|
158 |
return f"Error calling Cohere API: {e}"
|
159 |
|
160 |
+
|
161 |
# -------------------
|
162 |
# Local HF Path
|
163 |
# -------------------
|
|
|
168 |
"HUGGINGFACE_HUB_TOKEN (or HF_TOKEN) is not set. "
|
169 |
"Either set it, or provide COHERE_API_KEY to use Cohere's hosted API."
|
170 |
)
|
171 |
+
|
172 |
login(token=HF_TOKEN, add_to_git_credential=False)
|
173 |
+
|
174 |
dtype, device_map = pick_dtype_and_map()
|
175 |
tok = AutoTokenizer.from_pretrained(
|
176 |
+
MODEL_ID,
|
177 |
+
token=HF_TOKEN,
|
178 |
+
use_fast=True,
|
179 |
+
model_max_length=4096,
|
180 |
+
padding_side="left",
|
181 |
+
trust_remote_code=True,
|
182 |
)
|
183 |
mdl = AutoModelForCausalLM.from_pretrained(
|
184 |
+
MODEL_ID,
|
185 |
+
token=HF_TOKEN,
|
186 |
+
device_map=device_map,
|
187 |
+
low_cpu_mem_usage=True,
|
188 |
+
torch_dtype=dtype,
|
189 |
+
trust_remote_code=True,
|
190 |
)
|
191 |
if mdl.config.eos_token_id is None and tok.eos_token_id is not None:
|
192 |
mdl.config.eos_token_id = tok.eos_token_id
|
193 |
return mdl, tok
|
194 |
|
195 |
+
|
196 |
def build_inputs(tokenizer, message, history):
|
197 |
msgs = []
|
198 |
for u, a in (history or []):
|
|
|
203 |
msgs, tokenize=True, add_generation_prompt=True, return_tensors="pt"
|
204 |
)
|
205 |
|
206 |
+
|
207 |
def local_generate(model, tokenizer, input_ids, max_new_tokens=350):
|
208 |
input_ids = input_ids.to(model.device)
|
209 |
with torch.no_grad():
|
|
|
221 |
text = tokenizer.decode(gen_only, skip_special_tokens=True)
|
222 |
return text.strip()
|
223 |
|
224 |
+
|
225 |
# -------------------
|
226 |
# Chat callback (no header/meta in chat replies)
|
227 |
# -------------------
|
|
|
246 |
except Exception as e:
|
247 |
return f"Error during chat: {e}"
|
248 |
|
249 |
+
|
250 |
# -------------------
|
251 |
# Connection check (keeps header/meta)
|
252 |
# -------------------
|
|
|
272 |
except Exception as e:
|
273 |
return f"{header(user_tz=user_tz)}Connection Status: ❌ Error\nDetails: {e}"
|
274 |
|
275 |
+
|
276 |
# -------------------
|
277 |
# UI
|
278 |
# -------------------
|
279 |
with gr.Blocks(theme=gr.themes.Default()) as demo:
|
280 |
+
# Hold browser timezone (e.g., "America/Toronto")
|
281 |
user_tz_state = gr.State("")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
282 |
|
283 |
+
# On load, capture browser timezone via JS and store in user_tz_state
|
284 |
+
demo.load(
|
285 |
+
fn=lambda tz: tz, # echo the JS value
|
286 |
+
inputs=None,
|
287 |
+
outputs=[user_tz_state], # outputs must be a LIST
|
288 |
+
js="() => Intl.DateTimeFormat().resolvedOptions().timeZone"
|
289 |
+
)
|
290 |
|
291 |
+
gr.Markdown("# Medical Decision Support AI")
|
292 |
|
293 |
with gr.Row():
|
294 |
btn = gr.Button("Check Connection Status")
|
|
|
303 |
chat = gr.ChatInterface(
|
304 |
fn=chat_fn,
|
305 |
type="messages",
|
306 |
+
additional_inputs=[user_tz_state], # pass timezone into chat_fn (for future use)
|
307 |
description="A medical decision support system that provides healthcare-related information and decision making support.",
|
308 |
examples=[
|
309 |
["What are the symptoms of hypertension?", ""],
|
|
|
313 |
cache_examples=False,
|
314 |
)
|
315 |
|
316 |
+
# Wire timezone into the connection check as well
|
317 |
btn.click(fn=check_connection, inputs=user_tz_state, outputs=status)
|
318 |
|
319 |
if __name__ == "__main__":
|
320 |
demo.launch()
|
321 |
|
322 |
|
|
|
|
|
|