Spaces:
Running
Running
Commit
·
3cca1c2
1
Parent(s):
4a82770
Upd smartRAG
Browse files
memory.py
CHANGED
@@ -1,79 +1,118 @@
|
|
1 |
# memory.py
|
2 |
-
import re
|
|
|
|
|
3 |
import numpy as np
|
4 |
import faiss
|
5 |
-
from collections import defaultdict, deque
|
6 |
-
from typing import List
|
7 |
from sentence_transformers import SentenceTransformer
|
8 |
from google import genai # must be configured in app.py and imported globally
|
9 |
import logging
|
10 |
|
11 |
-
|
12 |
# Load embedding model
|
13 |
-
|
14 |
logger = logging.getLogger("medical-chatbot")
|
15 |
|
16 |
class MemoryManager:
|
17 |
-
def __init__(self, max_users=1000, history_per_user=10):
|
18 |
-
self.text_cache
|
19 |
-
self.chunk_index
|
20 |
-
self.
|
21 |
-
self.user_queue
|
|
|
|
|
22 |
|
|
|
23 |
def add_exchange(self, user_id: str, query: str, response: str, lang: str = "EN"):
|
24 |
-
|
25 |
-
if len(self.user_queue) >= self.user_queue.maxlen:
|
26 |
-
oldest = self.user_queue.popleft()
|
27 |
-
self._drop_user(oldest)
|
28 |
-
self.user_queue.append(user_id)
|
29 |
-
# Normalize
|
30 |
self.text_cache[user_id].append((query.strip(), response.strip()))
|
31 |
-
#
|
32 |
-
|
33 |
-
|
|
|
|
|
|
|
|
|
|
|
34 |
for chunk in chunks:
|
35 |
-
vec =
|
36 |
self.chunk_index[user_id].add(np.array([vec]))
|
37 |
-
self.
|
|
|
|
|
|
|
38 |
|
39 |
-
def get_relevant_chunks(self, user_id: str, query: str, top_k: int =
|
40 |
-
|
|
|
41 |
return []
|
42 |
-
# Encode
|
43 |
-
|
44 |
-
|
45 |
-
|
|
|
|
|
|
|
|
|
46 |
|
47 |
-
def get_context(self, user_id: str, num_turns: int = 3):
|
48 |
history = list(self.text_cache.get(user_id, []))[-num_turns:]
|
49 |
return "\n".join(f"User: {q}\nBot: {r}" for q, r in history)
|
50 |
|
51 |
def reset(self, user_id: str):
|
52 |
self._drop_user(user_id)
|
|
|
|
|
|
|
|
|
|
|
53 |
if user_id in self.user_queue:
|
54 |
self.user_queue.remove(user_id)
|
|
|
55 |
|
56 |
-
def _drop_user(self, user_id):
|
57 |
self.text_cache.pop(user_id, None)
|
58 |
self.chunk_index.pop(user_id, None)
|
59 |
-
self.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
|
61 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
"""
|
63 |
-
|
64 |
-
|
|
|
|
|
|
|
65 |
"""
|
66 |
-
#
|
67 |
instructions = []
|
68 |
-
# Only add translation if necessary
|
69 |
if lang.upper() != "EN":
|
70 |
instructions.append("- Translate the response to English.")
|
71 |
instructions.append("- Break the translated (or original) text into semantically distinct parts, grouped by medical topic or symptom.")
|
72 |
instructions.append("- For each part, generate a clear, concise summary. The summary may vary in length depending on the complexity of the topic — do not omit key clinical instructions.")
|
73 |
instructions.append("- Separate each part using three dashes `---` on a new line.")
|
74 |
-
#
|
75 |
-
joined_instructions = "\n".join(instructions)
|
76 |
-
# Prompting
|
77 |
prompt = f"""
|
78 |
You are a medical assistant helping organize and condense a clinical response.
|
79 |
Below is the user-provided medical response written in `{lang}`:
|
@@ -81,19 +120,33 @@ class MemoryManager:
|
|
81 |
{response}
|
82 |
------------------------
|
83 |
Please perform the following tasks:
|
84 |
-
{
|
|
|
85 |
Output only the structured summaries, separated by dashes.
|
86 |
"""
|
87 |
try:
|
88 |
client = genai.Client()
|
89 |
result = client.models.generate_content(
|
90 |
-
model=
|
91 |
contents=prompt,
|
92 |
generation_config={"temperature": 0.4}
|
93 |
)
|
94 |
output = result.text.strip()
|
95 |
-
logger.info(f"
|
96 |
-
return [
|
|
|
|
|
|
|
97 |
except Exception as e:
|
98 |
-
|
99 |
-
return [response.strip()]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
# memory.py
|
2 |
+
import re, time, hashlib, asyncio
|
3 |
+
from collections import defaultdict, deque
|
4 |
+
from typing import List, Dict
|
5 |
import numpy as np
|
6 |
import faiss
|
|
|
|
|
7 |
from sentence_transformers import SentenceTransformer
|
8 |
from google import genai # must be configured in app.py and imported globally
|
9 |
import logging
|
10 |
|
11 |
+
_LLM_SMALL = "gemini-2.5-flash-lite-preview-06-17"
|
12 |
# Load embedding model
|
13 |
+
EMBED = SentenceTransformer("/app/model_cache", device="cpu").half()
|
14 |
logger = logging.getLogger("medical-chatbot")
|
15 |
|
16 |
class MemoryManager:
|
17 |
+
def __init__(self, max_users=1000, history_per_user=10, max_chunks=30):
|
18 |
+
self.text_cache = defaultdict(lambda: deque(maxlen=history_per_user))
|
19 |
+
self.chunk_index = defaultdict(self._new_index) # user_id -> faiss index
|
20 |
+
self.chunk_meta = defaultdict(list) # '' -> list[{text,tag}]
|
21 |
+
self.user_queue = deque(maxlen=max_users) # LRU of users
|
22 |
+
self.max_chunks = max_chunks # hard cap per user
|
23 |
+
self.chunk_cache = {} # hash(query+resp) -> [chunks]
|
24 |
|
25 |
+
# ---------- Public API ----------
|
26 |
def add_exchange(self, user_id: str, query: str, response: str, lang: str = "EN"):
|
27 |
+
self._touch_user(user_id)
|
|
|
|
|
|
|
|
|
|
|
28 |
self.text_cache[user_id].append((query.strip(), response.strip()))
|
29 |
+
# Avoid re-chunking identical response
|
30 |
+
cache_key = hashlib.md5((query + response).encode()).hexdigest()
|
31 |
+
if cache_key in self.chunk_cache:
|
32 |
+
chunks = self.chunk_cache[cache_key]
|
33 |
+
else:
|
34 |
+
chunks = self._chunk_and_tag(response, lang)
|
35 |
+
self.chunk_cache[cache_key] = chunks
|
36 |
+
# Store chunks → faiss
|
37 |
for chunk in chunks:
|
38 |
+
vec = self._embed(chunk["text"])
|
39 |
self.chunk_index[user_id].add(np.array([vec]))
|
40 |
+
self.chunk_meta[user_id].append(chunk)
|
41 |
+
# Trim to max_chunks to keep latency O(1)
|
42 |
+
if len(self.chunk_meta[user_id]) > self.max_chunks:
|
43 |
+
self._rebuild_index(user_id, keep_last=self.max_chunks)
|
44 |
|
45 |
+
def get_relevant_chunks(self, user_id: str, query: str, top_k: int = 3, min_sim: float = 0.30) -> List[str]:
|
46 |
+
"""Return texts of chunks whose cosine similarity ≥ min_sim."""
|
47 |
+
if self.chunk_index[user_id].ntotal == 0:
|
48 |
return []
|
49 |
+
# Encode chunk
|
50 |
+
qvec = self._embed(query)
|
51 |
+
sims, idxs = self.chunk_index[user_id].search(np.array([qvec]), k=top_k)
|
52 |
+
results = []
|
53 |
+
for sim, idx in zip(sims[0], idxs[0]):
|
54 |
+
if idx < len(self.chunk_meta[user_id]) and sim >= min_sim:
|
55 |
+
results.append(self.chunk_meta[user_id][idx]["text"])
|
56 |
+
return results
|
57 |
|
58 |
+
def get_context(self, user_id: str, num_turns: int = 3) -> str:
|
59 |
history = list(self.text_cache.get(user_id, []))[-num_turns:]
|
60 |
return "\n".join(f"User: {q}\nBot: {r}" for q, r in history)
|
61 |
|
62 |
def reset(self, user_id: str):
|
63 |
self._drop_user(user_id)
|
64 |
+
|
65 |
+
# ---------- Internal helpers ----------
|
66 |
+
def _touch_user(self, user_id: str):
|
67 |
+
if user_id not in self.text_cache and len(self.user_queue) >= self.user_queue.maxlen:
|
68 |
+
self._drop_user(self.user_queue.popleft())
|
69 |
if user_id in self.user_queue:
|
70 |
self.user_queue.remove(user_id)
|
71 |
+
self.user_queue.append(user_id)
|
72 |
|
73 |
+
def _drop_user(self, user_id: str):
|
74 |
self.text_cache.pop(user_id, None)
|
75 |
self.chunk_index.pop(user_id, None)
|
76 |
+
self.chunk_meta.pop(user_id, None)
|
77 |
+
if user_id in self.user_queue:
|
78 |
+
self.user_queue.remove(user_id)
|
79 |
+
|
80 |
+
def _rebuild_index(self, user_id: str, keep_last: int):
|
81 |
+
"""Trim chunk list + rebuild FAISS index for user."""
|
82 |
+
self.chunk_meta[user_id] = self.chunk_meta[user_id][-keep_last:]
|
83 |
+
index = self._new_index()
|
84 |
+
for chunk in self.chunk_meta[user_id]:
|
85 |
+
vec = self._embed(chunk["text"])
|
86 |
+
index.add(np.array([vec]))
|
87 |
+
self.chunk_index[user_id] = index
|
88 |
+
|
89 |
+
@staticmethod
|
90 |
+
def _new_index():
|
91 |
+
# Use cosine similarity (vectors must be L2-normalised)
|
92 |
+
return faiss.IndexFlatIP(384)
|
93 |
|
94 |
+
@staticmethod
|
95 |
+
def _embed(text: str):
|
96 |
+
vec = EMBED.encode(text, convert_to_numpy=True)
|
97 |
+
# L2 normalise for cosine on IndexFlatIP
|
98 |
+
return vec / (np.linalg.norm(vec) + 1e-9)
|
99 |
+
|
100 |
+
def chunk_response(self, response: str, lang: str) -> List[Dict]:
|
101 |
"""
|
102 |
+
Calls Gemini to:
|
103 |
+
- Translate (if needed)
|
104 |
+
- Chunk by context/topic
|
105 |
+
- Summarise
|
106 |
+
Returns: [{"tag": ..., "text": ...}, ...]
|
107 |
"""
|
108 |
+
# Gemini instruction
|
109 |
instructions = []
|
|
|
110 |
if lang.upper() != "EN":
|
111 |
instructions.append("- Translate the response to English.")
|
112 |
instructions.append("- Break the translated (or original) text into semantically distinct parts, grouped by medical topic or symptom.")
|
113 |
instructions.append("- For each part, generate a clear, concise summary. The summary may vary in length depending on the complexity of the topic — do not omit key clinical instructions.")
|
114 |
instructions.append("- Separate each part using three dashes `---` on a new line.")
|
115 |
+
# Gemini prompt
|
|
|
|
|
116 |
prompt = f"""
|
117 |
You are a medical assistant helping organize and condense a clinical response.
|
118 |
Below is the user-provided medical response written in `{lang}`:
|
|
|
120 |
{response}
|
121 |
------------------------
|
122 |
Please perform the following tasks:
|
123 |
+
{chr(10).join(instructions)}
|
124 |
+
|
125 |
Output only the structured summaries, separated by dashes.
|
126 |
"""
|
127 |
try:
|
128 |
client = genai.Client()
|
129 |
result = client.models.generate_content(
|
130 |
+
model=_LLM_SMALL,
|
131 |
contents=prompt,
|
132 |
generation_config={"temperature": 0.4}
|
133 |
)
|
134 |
output = result.text.strip()
|
135 |
+
logger.info(f"📦 Gemini summarized chunk output: {output}")
|
136 |
+
return [
|
137 |
+
{"tag": self._quick_extract_topic(chunk), "text": chunk.strip()}
|
138 |
+
for chunk in output.split('---') if chunk.strip()
|
139 |
+
]
|
140 |
except Exception as e:
|
141 |
+
logger.warning(f"❌ Gemini chunking failed: {e}")
|
142 |
+
return [{"tag": "general", "text": response.strip()}]
|
143 |
+
|
144 |
+
@staticmethod
|
145 |
+
def _quick_extract_topic(chunk: str) -> str:
|
146 |
+
"""Heuristically extract the topic from a chunk (title line or first 3 words)."""
|
147 |
+
lines = chunk.strip().splitlines()
|
148 |
+
for line in lines:
|
149 |
+
if len(line.split()) <= 8 and line.strip().endswith(":"):
|
150 |
+
return line.strip().rstrip(":")
|
151 |
+
return " ".join(chunk.split()[:3]).rstrip(":.,")
|
152 |
+
|