Change tokenizer selection
Browse filesAdd system prompt
Add start message
Add incorporation of patient data
app.py
CHANGED
@@ -9,51 +9,6 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStream
|
|
9 |
import threading
|
10 |
import queue
|
11 |
|
12 |
-
class RichTextStreamer(TextIteratorStreamer):
|
13 |
-
def __init__(self, tokenizer, prompt_len=0, **kwargs):
|
14 |
-
super().__init__(tokenizer, **kwargs)
|
15 |
-
self.token_queue = queue.Queue()
|
16 |
-
self.prompt_len = prompt_len
|
17 |
-
self.count = 0
|
18 |
-
|
19 |
-
def put(self, value):
|
20 |
-
if isinstance(value, torch.Tensor):
|
21 |
-
token_ids = value.view(-1).tolist()
|
22 |
-
elif isinstance(value, list):
|
23 |
-
token_ids = value
|
24 |
-
else:
|
25 |
-
token_ids = [value]
|
26 |
-
|
27 |
-
for token_id in token_ids:
|
28 |
-
self.count += 1
|
29 |
-
if self.count <= self.prompt_len:
|
30 |
-
continue # skip prompt tokens
|
31 |
-
token_str = self.tokenizer.decode([token_id], **self.decode_kwargs)
|
32 |
-
is_special = token_id in self.tokenizer.all_special_ids
|
33 |
-
self.token_queue.put({
|
34 |
-
"token_id": token_id,
|
35 |
-
"token": token_str,
|
36 |
-
"is_special": is_special
|
37 |
-
})
|
38 |
-
|
39 |
-
def __iter__(self):
|
40 |
-
while True:
|
41 |
-
try:
|
42 |
-
token_info = self.token_queue.get(timeout=self.timeout)
|
43 |
-
yield token_info
|
44 |
-
except queue.Empty:
|
45 |
-
if self.end_of_generation.is_set():
|
46 |
-
break
|
47 |
-
|
48 |
-
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
|
49 |
-
import threading
|
50 |
-
|
51 |
-
from transformers import TextIteratorStreamer
|
52 |
-
import threading
|
53 |
-
|
54 |
-
from transformers import TextIteratorStreamer
|
55 |
-
import queue
|
56 |
-
|
57 |
class RichTextStreamer(TextIteratorStreamer):
|
58 |
def __init__(self, tokenizer, prompt_len=0, **kwargs):
|
59 |
super().__init__(tokenizer, **kwargs)
|
@@ -108,22 +63,66 @@ def chat_with_model(messages):
|
|
108 |
max_new_tokens = 1024
|
109 |
generated_tokens = 0
|
110 |
|
111 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
112 |
device = torch.device("cuda")
|
113 |
current_model.to(device).half()
|
114 |
|
115 |
-
# 1. Tokenize prompt
|
116 |
inputs = current_tokenizer(prompt, return_tensors="pt").to(device)
|
117 |
prompt_len = inputs["input_ids"].shape[-1]
|
118 |
|
119 |
-
# 2. Init streamer with prompt_len
|
120 |
streamer = RichTextStreamer(
|
121 |
tokenizer=current_tokenizer,
|
122 |
prompt_len=prompt_len,
|
123 |
skip_special_tokens=False
|
124 |
)
|
125 |
|
126 |
-
# 3. Build generation kwargs
|
127 |
generation_kwargs = dict(
|
128 |
**inputs,
|
129 |
max_new_tokens=max_new_tokens,
|
@@ -133,27 +132,20 @@ def chat_with_model(messages):
|
|
133 |
pad_token_id=pad_id
|
134 |
)
|
135 |
|
136 |
-
# 4. Launch generation in a thread
|
137 |
thread = threading.Thread(target=current_model.generate, kwargs=generation_kwargs)
|
138 |
thread.start()
|
139 |
|
140 |
messages = messages.copy()
|
141 |
messages.append({"role": "assistant", "content": ""})
|
142 |
|
143 |
-
print(f'Step 1: {messages}')
|
144 |
-
|
145 |
-
prompt_text = current_tokenizer.decode(inputs["input_ids"][0], skip_special_tokens=False)
|
146 |
-
|
147 |
for token_info in streamer:
|
148 |
token_str = token_info["token"]
|
149 |
token_id = token_info["token_id"]
|
150 |
is_special = token_info["is_special"]
|
151 |
|
152 |
-
# Stop immediately at EOS
|
153 |
if token_id == eos_id:
|
154 |
break
|
155 |
|
156 |
-
# Detect reasoning block
|
157 |
if "<think>" in token_str:
|
158 |
in_think = True
|
159 |
token_str = token_str.replace("<think>", "")
|
@@ -166,7 +158,6 @@ def chat_with_model(messages):
|
|
166 |
else:
|
167 |
output_text += token_str
|
168 |
|
169 |
-
# Early stopping if user reappears
|
170 |
if "\nUser" in output_text:
|
171 |
output_text = output_text.split("\nUser")[0].rstrip()
|
172 |
messages[-1]["content"] = output_text
|
@@ -178,34 +169,35 @@ def chat_with_model(messages):
|
|
178 |
|
179 |
messages[-1]["content"] = output_text
|
180 |
|
181 |
-
print(f'Step 2: {messages}')
|
182 |
-
|
183 |
yield messages
|
184 |
|
185 |
if in_think:
|
186 |
output_text += "*"
|
187 |
messages[-1]["content"] = output_text
|
188 |
-
|
189 |
-
# Wait for thread to finish
|
190 |
-
# current_model.to("cpu")
|
191 |
-
torch.cuda.empty_cache()
|
192 |
|
|
|
193 |
messages[-1]["content"] = output_text
|
194 |
-
print(f'Step 3: {messages}')
|
195 |
-
|
196 |
return messages
|
197 |
|
198 |
|
199 |
|
|
|
200 |
# Globals
|
201 |
current_model = None
|
202 |
current_tokenizer = None
|
203 |
|
|
|
|
|
204 |
def load_model_on_selection(model_name, progress=gr.Progress(track_tqdm=False)):
|
205 |
global current_model, current_tokenizer
|
206 |
token = os.getenv("HF_TOKEN")
|
207 |
|
208 |
-
progress(0, desc="Loading
|
|
|
|
|
|
|
|
|
|
|
209 |
current_tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token=token)
|
210 |
|
211 |
progress(0.5, desc="Loading model...")
|
@@ -219,6 +211,7 @@ def load_model_on_selection(model_name, progress=gr.Progress(track_tqdm=False)):
|
|
219 |
progress(1, desc="Model ready.")
|
220 |
return f"{model_name} loaded and ready!"
|
221 |
|
|
|
222 |
# Format conversation as plain text
|
223 |
def format_prompt(messages):
|
224 |
prompt = ""
|
@@ -239,7 +232,7 @@ model_choices = [
|
|
239 |
"meta-llama/Llama-3.2-3B-Instruct",
|
240 |
"deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
|
241 |
"google/gemma-7b",
|
242 |
-
"mistralai/Mistral-
|
243 |
]
|
244 |
|
245 |
# Example patient database
|
@@ -277,7 +270,7 @@ def autofill_patient(patient_key):
|
|
277 |
return "", "", "", ""
|
278 |
|
279 |
with gr.Blocks(css=".gradio-container {height: 100vh; overflow: hidden;}") as demo:
|
280 |
-
gr.Markdown("
|
281 |
|
282 |
default_model = gr.State(model_choices[0])
|
283 |
|
|
|
9 |
import threading
|
10 |
import queue
|
11 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
class RichTextStreamer(TextIteratorStreamer):
|
13 |
def __init__(self, tokenizer, prompt_len=0, **kwargs):
|
14 |
super().__init__(tokenizer, **kwargs)
|
|
|
63 |
max_new_tokens = 1024
|
64 |
generated_tokens = 0
|
65 |
|
66 |
+
# ⮕ PREPARE SYSTEM + INITIAL MESSAGES
|
67 |
+
system_messages = [
|
68 |
+
{
|
69 |
+
"role": "system",
|
70 |
+
"content": (
|
71 |
+
"You are a radiologist's companion, here to answer questions about the patient and assist in the diagnosis if asked to do so. "
|
72 |
+
"You are able to call specialized tools. "
|
73 |
+
"At the moment, you have one tool available: an organ segmentation algorithm for abdominal CTs.\n\n"
|
74 |
+
"If the user requests an organ segmentation, output a JSON object in this structure:\n"
|
75 |
+
"{\n"
|
76 |
+
" \"function\": \"segment_organ\",\n"
|
77 |
+
" \"arguments\": {\n"
|
78 |
+
" \"scan_path\": \"<path_to_ct_scan>\",\n"
|
79 |
+
" \"organ\": \"<organ_name>\"\n"
|
80 |
+
" }\n"
|
81 |
+
"}\n\n"
|
82 |
+
"Once you call the function, the app will execute it and return the result."
|
83 |
+
)
|
84 |
+
},
|
85 |
+
{
|
86 |
+
"role": "system",
|
87 |
+
"content": f"Patient Information:\nName: {patient_name.value}\nAge: {patient_age.value}\nID: {patient_id.value}\nNotes: {patient_notes.value}"
|
88 |
+
}
|
89 |
+
]
|
90 |
+
|
91 |
+
# Optional: if you later add available_images, you could append another system message.
|
92 |
+
|
93 |
+
welcome_message = (
|
94 |
+
"**Welcome to the Radiologist's Companion!**\n\n"
|
95 |
+
"You can ask me about the patient's medical history or available imaging data.\n"
|
96 |
+
"- I can summarize key details from the EHR.\n"
|
97 |
+
"- I can tell you which medical images are available.\n"
|
98 |
+
"- If you'd like an organ segmentation (e.g. spleen, liver, kidney_left, colon, femur_right) on an abdominal CT scan, just ask!\n\n"
|
99 |
+
"**Example Requests:**\n"
|
100 |
+
"- \"What do we know about this patient?\"\n"
|
101 |
+
"- \"Which images are available for this patient?\"\n"
|
102 |
+
"- \"Can you segment the spleen from the CT scan?\"\n"
|
103 |
+
)
|
104 |
+
|
105 |
+
# If it's the first user message (i.e., no assistant yet), prepend welcome
|
106 |
+
if len(messages) == 1 and messages[0]['role'] == 'user':
|
107 |
+
messages = [{"role": "assistant", "content": welcome_message}] + messages
|
108 |
+
|
109 |
+
# Merge full conversation
|
110 |
+
full_messages = system_messages + messages
|
111 |
+
|
112 |
+
prompt = format_prompt(full_messages)
|
113 |
+
|
114 |
device = torch.device("cuda")
|
115 |
current_model.to(device).half()
|
116 |
|
|
|
117 |
inputs = current_tokenizer(prompt, return_tensors="pt").to(device)
|
118 |
prompt_len = inputs["input_ids"].shape[-1]
|
119 |
|
|
|
120 |
streamer = RichTextStreamer(
|
121 |
tokenizer=current_tokenizer,
|
122 |
prompt_len=prompt_len,
|
123 |
skip_special_tokens=False
|
124 |
)
|
125 |
|
|
|
126 |
generation_kwargs = dict(
|
127 |
**inputs,
|
128 |
max_new_tokens=max_new_tokens,
|
|
|
132 |
pad_token_id=pad_id
|
133 |
)
|
134 |
|
|
|
135 |
thread = threading.Thread(target=current_model.generate, kwargs=generation_kwargs)
|
136 |
thread.start()
|
137 |
|
138 |
messages = messages.copy()
|
139 |
messages.append({"role": "assistant", "content": ""})
|
140 |
|
|
|
|
|
|
|
|
|
141 |
for token_info in streamer:
|
142 |
token_str = token_info["token"]
|
143 |
token_id = token_info["token_id"]
|
144 |
is_special = token_info["is_special"]
|
145 |
|
|
|
146 |
if token_id == eos_id:
|
147 |
break
|
148 |
|
|
|
149 |
if "<think>" in token_str:
|
150 |
in_think = True
|
151 |
token_str = token_str.replace("<think>", "")
|
|
|
158 |
else:
|
159 |
output_text += token_str
|
160 |
|
|
|
161 |
if "\nUser" in output_text:
|
162 |
output_text = output_text.split("\nUser")[0].rstrip()
|
163 |
messages[-1]["content"] = output_text
|
|
|
169 |
|
170 |
messages[-1]["content"] = output_text
|
171 |
|
|
|
|
|
172 |
yield messages
|
173 |
|
174 |
if in_think:
|
175 |
output_text += "*"
|
176 |
messages[-1]["content"] = output_text
|
|
|
|
|
|
|
|
|
177 |
|
178 |
+
torch.cuda.empty_cache()
|
179 |
messages[-1]["content"] = output_text
|
|
|
|
|
180 |
return messages
|
181 |
|
182 |
|
183 |
|
184 |
+
|
185 |
# Globals
|
186 |
current_model = None
|
187 |
current_tokenizer = None
|
188 |
|
189 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, AutoConfig, LlamaTokenizer
|
190 |
+
|
191 |
def load_model_on_selection(model_name, progress=gr.Progress(track_tqdm=False)):
|
192 |
global current_model, current_tokenizer
|
193 |
token = os.getenv("HF_TOKEN")
|
194 |
|
195 |
+
progress(0, desc="Loading config...")
|
196 |
+
config = AutoConfig.from_pretrained(model_name, use_auth_token=token)
|
197 |
+
|
198 |
+
progress(0.2, desc="Loading tokenizer...")
|
199 |
+
|
200 |
+
# Default
|
201 |
current_tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token=token)
|
202 |
|
203 |
progress(0.5, desc="Loading model...")
|
|
|
211 |
progress(1, desc="Model ready.")
|
212 |
return f"{model_name} loaded and ready!"
|
213 |
|
214 |
+
|
215 |
# Format conversation as plain text
|
216 |
def format_prompt(messages):
|
217 |
prompt = ""
|
|
|
232 |
"meta-llama/Llama-3.2-3B-Instruct",
|
233 |
"deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
|
234 |
"google/gemma-7b",
|
235 |
+
"mistralai/Mistral-Nemo-Instruct-FP8-2407"
|
236 |
]
|
237 |
|
238 |
# Example patient database
|
|
|
270 |
return "", "", "", ""
|
271 |
|
272 |
with gr.Blocks(css=".gradio-container {height: 100vh; overflow: hidden;}") as demo:
|
273 |
+
gr.Markdown("<h2 style='text-align: center;'>Radiologist's Companion</h2>")
|
274 |
|
275 |
default_model = gr.State(model_choices[0])
|
276 |
|