llms-py 3.0.0b7__py3-none-any.whl → 3.0.0b9__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- llms/__pycache__/main.cpython-314.pyc +0 -0
- llms/extensions/analytics/ui/index.mjs +51 -162
- llms/extensions/app/__init__.py +519 -0
- llms/extensions/app/__pycache__/__init__.cpython-314.pyc +0 -0
- llms/extensions/app/__pycache__/db.cpython-314.pyc +0 -0
- llms/extensions/app/__pycache__/db_manager.cpython-314.pyc +0 -0
- llms/extensions/app/db.py +643 -0
- llms/extensions/app/db_manager.py +195 -0
- llms/extensions/app/requests.json +9073 -0
- llms/extensions/app/threads.json +15290 -0
- llms/{ui/modules/threads → extensions/app/ui}/Recents.mjs +82 -55
- llms/{ui/modules/threads → extensions/app/ui}/index.mjs +78 -9
- llms/extensions/app/ui/threadStore.mjs +407 -0
- llms/extensions/core_tools/__init__.py +272 -32
- llms/extensions/core_tools/__pycache__/__init__.cpython-314.pyc +0 -0
- llms/extensions/core_tools/ui/codemirror/addon/edit/closebrackets.js +201 -0
- llms/extensions/core_tools/ui/codemirror/addon/edit/closetag.js +185 -0
- llms/extensions/core_tools/ui/codemirror/addon/edit/continuelist.js +101 -0
- llms/extensions/core_tools/ui/codemirror/addon/edit/matchbrackets.js +160 -0
- llms/extensions/core_tools/ui/codemirror/addon/edit/matchtags.js +66 -0
- llms/extensions/core_tools/ui/codemirror/addon/edit/trailingspace.js +27 -0
- llms/extensions/core_tools/ui/codemirror/addon/selection/active-line.js +72 -0
- llms/extensions/core_tools/ui/codemirror/addon/selection/mark-selection.js +119 -0
- llms/extensions/core_tools/ui/codemirror/addon/selection/selection-pointer.js +98 -0
- llms/extensions/core_tools/ui/codemirror/doc/docs.css +225 -0
- llms/extensions/core_tools/ui/codemirror/doc/source_sans.woff +0 -0
- llms/extensions/core_tools/ui/codemirror/lib/codemirror.css +344 -0
- llms/extensions/core_tools/ui/codemirror/lib/codemirror.js +9884 -0
- llms/extensions/core_tools/ui/codemirror/mode/clike/clike.js +942 -0
- llms/extensions/core_tools/ui/codemirror/mode/javascript/index.html +118 -0
- llms/extensions/core_tools/ui/codemirror/mode/javascript/javascript.js +962 -0
- llms/extensions/core_tools/ui/codemirror/mode/javascript/typescript.html +62 -0
- llms/extensions/core_tools/ui/codemirror/mode/python/python.js +402 -0
- llms/extensions/core_tools/ui/codemirror/theme/dracula.css +40 -0
- llms/extensions/core_tools/ui/codemirror/theme/mocha.css +135 -0
- llms/extensions/core_tools/ui/index.mjs +650 -0
- llms/extensions/gallery/__pycache__/db.cpython-314.pyc +0 -0
- llms/extensions/gallery/db.py +4 -4
- llms/extensions/gallery/ui/index.mjs +2 -1
- llms/extensions/katex/__init__.py +6 -0
- llms/extensions/katex/__pycache__/__init__.cpython-314.pyc +0 -0
- llms/extensions/katex/ui/README.md +125 -0
- llms/extensions/katex/ui/contrib/auto-render.js +338 -0
- llms/extensions/katex/ui/contrib/auto-render.min.js +1 -0
- llms/extensions/katex/ui/contrib/auto-render.mjs +244 -0
- llms/extensions/katex/ui/contrib/copy-tex.js +127 -0
- llms/extensions/katex/ui/contrib/copy-tex.min.js +1 -0
- llms/extensions/katex/ui/contrib/copy-tex.mjs +105 -0
- llms/extensions/katex/ui/contrib/mathtex-script-type.js +109 -0
- llms/extensions/katex/ui/contrib/mathtex-script-type.min.js +1 -0
- llms/extensions/katex/ui/contrib/mathtex-script-type.mjs +24 -0
- llms/extensions/katex/ui/contrib/mhchem.js +3213 -0
- llms/extensions/katex/ui/contrib/mhchem.min.js +1 -0
- llms/extensions/katex/ui/contrib/mhchem.mjs +3109 -0
- llms/extensions/katex/ui/contrib/render-a11y-string.js +887 -0
- llms/extensions/katex/ui/contrib/render-a11y-string.min.js +1 -0
- llms/extensions/katex/ui/contrib/render-a11y-string.mjs +800 -0
- llms/extensions/katex/ui/fonts/KaTeX_AMS-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_AMS-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_AMS-Regular.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Caligraphic-Bold.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Caligraphic-Bold.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Caligraphic-Bold.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Caligraphic-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Caligraphic-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Caligraphic-Regular.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Fraktur-Bold.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Fraktur-Bold.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Fraktur-Bold.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Fraktur-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Fraktur-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Fraktur-Regular.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-Bold.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-Bold.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-Bold.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-BoldItalic.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-BoldItalic.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-BoldItalic.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-Italic.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-Italic.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-Italic.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Main-Regular.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Math-BoldItalic.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Math-BoldItalic.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Math-BoldItalic.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Math-Italic.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Math-Italic.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Math-Italic.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_SansSerif-Bold.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_SansSerif-Bold.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_SansSerif-Bold.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_SansSerif-Italic.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_SansSerif-Italic.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_SansSerif-Italic.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_SansSerif-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_SansSerif-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_SansSerif-Regular.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Script-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Script-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Script-Regular.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size1-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size1-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size1-Regular.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size2-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size2-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size2-Regular.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size3-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size3-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size3-Regular.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size4-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size4-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Size4-Regular.woff2 +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Typewriter-Regular.ttf +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Typewriter-Regular.woff +0 -0
- llms/extensions/katex/ui/fonts/KaTeX_Typewriter-Regular.woff2 +0 -0
- llms/extensions/katex/ui/index.mjs +92 -0
- llms/extensions/katex/ui/katex-swap.css +1230 -0
- llms/extensions/katex/ui/katex-swap.min.css +1 -0
- llms/extensions/katex/ui/katex.css +1230 -0
- llms/extensions/katex/ui/katex.js +19080 -0
- llms/extensions/katex/ui/katex.min.css +1 -0
- llms/extensions/katex/ui/katex.min.js +1 -0
- llms/extensions/katex/ui/katex.min.mjs +1 -0
- llms/extensions/katex/ui/katex.mjs +18547 -0
- llms/extensions/providers/__pycache__/anthropic.cpython-314.pyc +0 -0
- llms/extensions/providers/anthropic.py +44 -1
- llms/extensions/system_prompts/ui/index.mjs +2 -1
- llms/extensions/tools/__init__.py +5 -0
- llms/extensions/tools/__pycache__/__init__.cpython-314.pyc +0 -0
- llms/extensions/tools/ui/index.mjs +8 -8
- llms/index.html +26 -38
- llms/llms.json +4 -1
- llms/main.py +492 -103
- llms/ui/App.mjs +2 -3
- llms/ui/ai.mjs +29 -13
- llms/ui/app.css +255 -289
- llms/ui/ctx.mjs +84 -6
- llms/ui/index.mjs +4 -6
- llms/ui/lib/vue.min.mjs +10 -9
- llms/ui/lib/vue.mjs +1796 -1635
- llms/ui/markdown.mjs +4 -2
- llms/ui/modules/chat/ChatBody.mjs +90 -86
- llms/ui/modules/chat/HomeTools.mjs +0 -242
- llms/ui/modules/chat/index.mjs +103 -170
- llms/ui/modules/model-selector.mjs +2 -2
- llms/ui/tailwind.input.css +35 -1
- llms/ui/utils.mjs +12 -0
- {llms_py-3.0.0b7.dist-info → llms_py-3.0.0b9.dist-info}/METADATA +1 -1
- llms_py-3.0.0b9.dist-info/RECORD +198 -0
- llms/ui/modules/threads/threadStore.mjs +0 -640
- llms_py-3.0.0b7.dist-info/RECORD +0 -80
- {llms_py-3.0.0b7.dist-info → llms_py-3.0.0b9.dist-info}/WHEEL +0 -0
- {llms_py-3.0.0b7.dist-info → llms_py-3.0.0b9.dist-info}/entry_points.txt +0 -0
- {llms_py-3.0.0b7.dist-info → llms_py-3.0.0b9.dist-info}/licenses/LICENSE +0 -0
- {llms_py-3.0.0b7.dist-info → llms_py-3.0.0b9.dist-info}/top_level.txt +0 -0
|
@@ -0,0 +1,519 @@
|
|
|
1
|
+
import asyncio
|
|
2
|
+
import json
|
|
3
|
+
import os
|
|
4
|
+
import time
|
|
5
|
+
from datetime import datetime
|
|
6
|
+
from typing import Any
|
|
7
|
+
|
|
8
|
+
from aiohttp import web
|
|
9
|
+
|
|
10
|
+
g_db = None
|
|
11
|
+
|
|
12
|
+
try:
|
|
13
|
+
from llms.extensions.app.db import AppDB
|
|
14
|
+
except ImportError as e:
|
|
15
|
+
print(f"Failed to import AppDB: {e}")
|
|
16
|
+
AppDB = None
|
|
17
|
+
|
|
18
|
+
|
|
19
|
+
def install(ctx):
|
|
20
|
+
def get_db():
|
|
21
|
+
global g_db
|
|
22
|
+
if g_db is None and AppDB:
|
|
23
|
+
try:
|
|
24
|
+
db_path = os.path.join(ctx.get_user_path(), "app", "app.sqlite")
|
|
25
|
+
g_db = AppDB(ctx, db_path)
|
|
26
|
+
ctx.register_shutdown_handler(g_db.close)
|
|
27
|
+
|
|
28
|
+
threads_json = "/home/mythz/src/ServiceStack/llms/llms/extensions/app/threads.json"
|
|
29
|
+
requests_json = "/home/mythz/src/ServiceStack/llms/llms/extensions/app/requests.json"
|
|
30
|
+
with open(threads_json) as f:
|
|
31
|
+
threads = json.load(f)["threads"]
|
|
32
|
+
threads.reverse()
|
|
33
|
+
with open(requests_json) as f:
|
|
34
|
+
requests = json.load(f)["requests"]
|
|
35
|
+
requests.reverse()
|
|
36
|
+
# g_db.import_db(threads, requests)
|
|
37
|
+
|
|
38
|
+
except Exception as e:
|
|
39
|
+
ctx.err("Failed to init AppDB", e)
|
|
40
|
+
return g_db
|
|
41
|
+
|
|
42
|
+
if not get_db():
|
|
43
|
+
return
|
|
44
|
+
|
|
45
|
+
def to_dto(row, json_columns):
|
|
46
|
+
# as=column -> [0,1,2]
|
|
47
|
+
if not isinstance(row, dict):
|
|
48
|
+
return row
|
|
49
|
+
|
|
50
|
+
to = {}
|
|
51
|
+
for k, v in row.items():
|
|
52
|
+
if k in json_columns and v is not None and isinstance(v, str):
|
|
53
|
+
try:
|
|
54
|
+
to[k] = json.loads(v)
|
|
55
|
+
except Exception as e:
|
|
56
|
+
ctx.err(f"Failed to parse JSON for {k}: {v} ({type(v)})", e)
|
|
57
|
+
to[k] = v
|
|
58
|
+
else:
|
|
59
|
+
to[k] = v
|
|
60
|
+
return to
|
|
61
|
+
|
|
62
|
+
def thread_dto(row):
|
|
63
|
+
return row and to_dto(row, ["messages", "modalities", "args", "modelInfo", "stats"])
|
|
64
|
+
|
|
65
|
+
def request_dto(row):
|
|
66
|
+
return row and to_dto(row, ["usage"])
|
|
67
|
+
|
|
68
|
+
def prompt_to_title(prompt):
|
|
69
|
+
return prompt[:100] + ("..." if len(prompt) > 100 else "") if prompt else None
|
|
70
|
+
|
|
71
|
+
def timestamp_messages(messages):
|
|
72
|
+
timestamp = int(time.time() * 1000)
|
|
73
|
+
for message in messages:
|
|
74
|
+
if "timestamp" not in message:
|
|
75
|
+
message["timestamp"] = timestamp
|
|
76
|
+
timestamp += 1 # make unique
|
|
77
|
+
return messages
|
|
78
|
+
|
|
79
|
+
async def query_threads(request):
|
|
80
|
+
rows = g_db.query_threads(request.query, user=ctx.get_username(request))
|
|
81
|
+
dtos = [thread_dto(row) for row in rows]
|
|
82
|
+
return web.json_response(dtos)
|
|
83
|
+
|
|
84
|
+
ctx.add_get("threads", query_threads)
|
|
85
|
+
|
|
86
|
+
async def create_thread(request):
|
|
87
|
+
thread = await request.json()
|
|
88
|
+
id = await g_db.create_thread_async(thread, user=ctx.get_username(request))
|
|
89
|
+
row = g_db.get_thread(id, user=ctx.get_username(request))
|
|
90
|
+
return web.json_response(thread_dto(row) if row else "")
|
|
91
|
+
|
|
92
|
+
ctx.add_post("threads", create_thread)
|
|
93
|
+
|
|
94
|
+
async def update_thread(request):
|
|
95
|
+
thread = await request.json()
|
|
96
|
+
id = request.match_info["id"]
|
|
97
|
+
update_count = await g_db.update_thread_async(id, thread, user=ctx.get_username(request))
|
|
98
|
+
if update_count == 0:
|
|
99
|
+
raise Exception("Thread not found")
|
|
100
|
+
row = g_db.get_thread(id, user=ctx.get_username(request))
|
|
101
|
+
return web.json_response(thread_dto(row) if row else "")
|
|
102
|
+
|
|
103
|
+
ctx.add_patch("threads/{id}", update_thread)
|
|
104
|
+
|
|
105
|
+
async def delete_thread(request):
|
|
106
|
+
id = request.match_info["id"]
|
|
107
|
+
g_db.delete_thread(id, user=ctx.get_username(request))
|
|
108
|
+
return web.json_response({})
|
|
109
|
+
|
|
110
|
+
ctx.add_delete("threads/{id}", delete_thread)
|
|
111
|
+
|
|
112
|
+
async def queue_chat_handler(request):
|
|
113
|
+
# Check authentication if enabled
|
|
114
|
+
is_authenticated, user_data = ctx.check_auth(request)
|
|
115
|
+
if not is_authenticated:
|
|
116
|
+
return web.json_response(ctx.error_auth_required, status=401)
|
|
117
|
+
|
|
118
|
+
if not request.body_exists:
|
|
119
|
+
raise Exception("messages required")
|
|
120
|
+
|
|
121
|
+
chat = await request.json()
|
|
122
|
+
|
|
123
|
+
messages = timestamp_messages(chat.get("messages", []))
|
|
124
|
+
if len(messages) == 0:
|
|
125
|
+
raise Exception("messages required")
|
|
126
|
+
|
|
127
|
+
id = request.match_info["id"]
|
|
128
|
+
thread = thread_dto(g_db.get_thread(id, user=ctx.get_username(request)))
|
|
129
|
+
if not thread:
|
|
130
|
+
raise Exception("Thread not found")
|
|
131
|
+
|
|
132
|
+
update_thread = {
|
|
133
|
+
"messages": messages,
|
|
134
|
+
"startedAt": datetime.now(),
|
|
135
|
+
"completedAt": None,
|
|
136
|
+
"error": None,
|
|
137
|
+
}
|
|
138
|
+
|
|
139
|
+
model = chat.get("model", None)
|
|
140
|
+
if model:
|
|
141
|
+
update_thread["model"] = model
|
|
142
|
+
metadata = chat.get("metadata", {})
|
|
143
|
+
if len(metadata) > 0:
|
|
144
|
+
update_thread["metadata"] = metadata
|
|
145
|
+
if chat.get("modalities") or not thread.get("modalities"):
|
|
146
|
+
update_thread["modalities"] = chat.get("modalities", ["text"])
|
|
147
|
+
system_prompt = ctx.chat_to_system_prompt(chat)
|
|
148
|
+
if system_prompt:
|
|
149
|
+
update_thread["systemPrompt"] = system_prompt
|
|
150
|
+
|
|
151
|
+
args = thread.get("args") or {}
|
|
152
|
+
for k, v in chat.items():
|
|
153
|
+
if k in ctx.request_args:
|
|
154
|
+
args[k] = v
|
|
155
|
+
update_thread["args"] = args
|
|
156
|
+
|
|
157
|
+
# allow chat to override thread title
|
|
158
|
+
title = chat.get("title")
|
|
159
|
+
if title:
|
|
160
|
+
update_thread["title"] = title
|
|
161
|
+
else:
|
|
162
|
+
# only update thread title if it's not already set
|
|
163
|
+
title = thread.get("title")
|
|
164
|
+
if not title:
|
|
165
|
+
update_thread["title"] = title = prompt_to_title(ctx.last_user_prompt(chat))
|
|
166
|
+
|
|
167
|
+
user = ctx.get_username(request)
|
|
168
|
+
await g_db.update_thread_async(
|
|
169
|
+
id,
|
|
170
|
+
update_thread,
|
|
171
|
+
user=user,
|
|
172
|
+
)
|
|
173
|
+
thread = thread_dto(g_db.get_thread(id, user=user))
|
|
174
|
+
if not thread:
|
|
175
|
+
raise Exception("Thread not found")
|
|
176
|
+
|
|
177
|
+
chat = {
|
|
178
|
+
"model": thread.get("model"),
|
|
179
|
+
"messages": thread.get("messages"),
|
|
180
|
+
"modalities": thread.get("modalities"),
|
|
181
|
+
"systemPrompt": thread.get("systemPrompt"),
|
|
182
|
+
"metadata": thread.get("metadata", {}),
|
|
183
|
+
}
|
|
184
|
+
for k, v in thread.get("args", {}).items():
|
|
185
|
+
if k in ctx.request_args:
|
|
186
|
+
chat[k] = v
|
|
187
|
+
|
|
188
|
+
context = {
|
|
189
|
+
"chat": chat,
|
|
190
|
+
"user": user,
|
|
191
|
+
"threadId": id,
|
|
192
|
+
"tools": chat.get("metadata").get("tools", "all"),
|
|
193
|
+
}
|
|
194
|
+
|
|
195
|
+
# execute chat in background thread
|
|
196
|
+
async def run_chat(chat_req, context_req):
|
|
197
|
+
try:
|
|
198
|
+
await ctx.chat_completion(chat_req, context=context_req)
|
|
199
|
+
except Exception as ex:
|
|
200
|
+
ctx.err("run_chat", ex)
|
|
201
|
+
# not necessary to update thread in db with error as it's done in chat_error filter
|
|
202
|
+
|
|
203
|
+
asyncio.create_task(run_chat(chat, context))
|
|
204
|
+
|
|
205
|
+
return web.json_response(thread_dto(thread))
|
|
206
|
+
|
|
207
|
+
ctx.add_post("threads/{id}/chat", queue_chat_handler)
|
|
208
|
+
|
|
209
|
+
async def get_thread_updates(request):
|
|
210
|
+
id = request.match_info["id"]
|
|
211
|
+
after = request.query.get("after", None)
|
|
212
|
+
user = ctx.get_username(request)
|
|
213
|
+
thread = g_db.get_thread(id, user=user)
|
|
214
|
+
if not thread:
|
|
215
|
+
raise Exception("Thread not found")
|
|
216
|
+
if after:
|
|
217
|
+
started = time.time()
|
|
218
|
+
thread_id = thread.get("id")
|
|
219
|
+
thread_updated_at = thread.get("updatedAt")
|
|
220
|
+
|
|
221
|
+
while thread_updated_at <= after:
|
|
222
|
+
thread_updated_at = g_db.get_thread_column(thread_id, "updatedAt", user=user)
|
|
223
|
+
# if thread is not updated in 30 seconds, break
|
|
224
|
+
if time.time() - started > 10:
|
|
225
|
+
break
|
|
226
|
+
await asyncio.sleep(1)
|
|
227
|
+
ctx.dbg(f"get_thread_updates: {thread_id} / {thread_updated_at} < {after} / {thread_updated_at < after}")
|
|
228
|
+
thread = g_db.get_thread(thread_id, user=user)
|
|
229
|
+
return web.json_response(thread_dto(thread))
|
|
230
|
+
|
|
231
|
+
ctx.add_get("threads/{id}/updates", get_thread_updates)
|
|
232
|
+
|
|
233
|
+
async def cancel_thread(request):
|
|
234
|
+
id = request.match_info["id"]
|
|
235
|
+
await g_db.update_thread_async(
|
|
236
|
+
id, {"completedAt": datetime.now(), "error": "Request was canceled"}, user=ctx.get_username(request)
|
|
237
|
+
)
|
|
238
|
+
thread = g_db.get_thread(id, user=ctx.get_username(request))
|
|
239
|
+
ctx.dbg(f"cancel_thread: {id} / {thread.get('error')} / {thread.get('completedAt')}")
|
|
240
|
+
return web.json_response(thread_dto(thread))
|
|
241
|
+
|
|
242
|
+
ctx.add_post("threads/{id}/cancel", cancel_thread)
|
|
243
|
+
|
|
244
|
+
async def query_requests(request):
|
|
245
|
+
rows = g_db.query_requests(request.query, user=ctx.get_username(request))
|
|
246
|
+
dtos = [request_dto(row) for row in rows]
|
|
247
|
+
return web.json_response(dtos)
|
|
248
|
+
|
|
249
|
+
ctx.add_get("requests", query_requests)
|
|
250
|
+
|
|
251
|
+
async def delete_request(request):
|
|
252
|
+
id = request.match_info["id"]
|
|
253
|
+
g_db.delete_request(id, user=ctx.get_username(request))
|
|
254
|
+
return web.json_response({})
|
|
255
|
+
|
|
256
|
+
ctx.add_delete("requests/{id}", delete_request)
|
|
257
|
+
|
|
258
|
+
async def requests_summary(request):
|
|
259
|
+
rows = g_db.get_request_summary(user=ctx.get_username(request))
|
|
260
|
+
stats = {
|
|
261
|
+
"dailyData": {},
|
|
262
|
+
"years": [],
|
|
263
|
+
"totalCost": 0,
|
|
264
|
+
"totalRequests": 0,
|
|
265
|
+
"totalInputTokens": 0,
|
|
266
|
+
"totalOutputTokens": 0,
|
|
267
|
+
}
|
|
268
|
+
years = set()
|
|
269
|
+
for row in rows:
|
|
270
|
+
date = row["date"]
|
|
271
|
+
year = int(date[:4])
|
|
272
|
+
years.add(year)
|
|
273
|
+
stats["dailyData"][date] = {
|
|
274
|
+
"cost": row["cost"],
|
|
275
|
+
"requests": row["requests"],
|
|
276
|
+
"inputTokens": row["inputTokens"],
|
|
277
|
+
"outputTokens": row["outputTokens"],
|
|
278
|
+
}
|
|
279
|
+
stats["totalCost"] += row["cost"] or 0
|
|
280
|
+
stats["totalRequests"] += row["requests"] or 0
|
|
281
|
+
stats["totalInputTokens"] += row["inputTokens"] or 0
|
|
282
|
+
stats["totalOutputTokens"] += row["outputTokens"] or 0
|
|
283
|
+
|
|
284
|
+
stats["years"] = sorted(years)
|
|
285
|
+
return web.json_response(stats)
|
|
286
|
+
|
|
287
|
+
ctx.add_get("requests/summary", requests_summary)
|
|
288
|
+
|
|
289
|
+
async def daily_requests_summary(request):
|
|
290
|
+
day = request.match_info["day"]
|
|
291
|
+
summary = g_db.get_daily_request_summary(day, user=ctx.get_username(request))
|
|
292
|
+
return web.json_response(summary)
|
|
293
|
+
|
|
294
|
+
ctx.add_get("requests/summary/{day}", daily_requests_summary)
|
|
295
|
+
|
|
296
|
+
async def chat_request(openai_request, context):
|
|
297
|
+
chat = openai_request
|
|
298
|
+
user = context.get("user", None)
|
|
299
|
+
provider = context.get("provider", None)
|
|
300
|
+
thread_id = context.get("threadId", None)
|
|
301
|
+
model_info = context.get("modelInfo", None)
|
|
302
|
+
|
|
303
|
+
metadata = chat.get("metadata", {})
|
|
304
|
+
model = chat.get("model", None)
|
|
305
|
+
messages = timestamp_messages(chat.get("messages", []))
|
|
306
|
+
title = context.get("title") or prompt_to_title(ctx.last_user_prompt(chat) if chat else None)
|
|
307
|
+
started_at = context.get("startedAt")
|
|
308
|
+
if not started_at:
|
|
309
|
+
context["startedAt"] = started_at = datetime.now()
|
|
310
|
+
if thread_id is None:
|
|
311
|
+
thread = {
|
|
312
|
+
"user": user,
|
|
313
|
+
"model": model,
|
|
314
|
+
"provider": provider,
|
|
315
|
+
"modelInfo": model_info,
|
|
316
|
+
"title": title,
|
|
317
|
+
"messages": messages,
|
|
318
|
+
"systemPrompt": ctx.chat_to_system_prompt(chat),
|
|
319
|
+
"modalities": chat.get("modalities", ["text"]),
|
|
320
|
+
"startedAt": started_at,
|
|
321
|
+
"metadata": metadata,
|
|
322
|
+
}
|
|
323
|
+
thread_id = await g_db.create_thread_async(thread, user=user)
|
|
324
|
+
context["threadId"] = thread_id
|
|
325
|
+
else:
|
|
326
|
+
update_thread = {
|
|
327
|
+
"model": model,
|
|
328
|
+
"provider": provider,
|
|
329
|
+
"modelInfo": model_info,
|
|
330
|
+
"startedAt": started_at,
|
|
331
|
+
"messages": messages,
|
|
332
|
+
"completedAt": None,
|
|
333
|
+
"error": None,
|
|
334
|
+
"metadata": metadata,
|
|
335
|
+
}
|
|
336
|
+
await g_db.update_thread_async(thread_id, update_thread, user=user)
|
|
337
|
+
|
|
338
|
+
completed_at = g_db.get_thread_column(thread_id, "completedAt", user=user)
|
|
339
|
+
if completed_at:
|
|
340
|
+
context["completed"] = True
|
|
341
|
+
|
|
342
|
+
ctx.register_chat_request_filter(chat_request)
|
|
343
|
+
|
|
344
|
+
async def tool_request(tool_chat, context):
|
|
345
|
+
ctx.dbg("tool_request")
|
|
346
|
+
thread_id = context.get("threadId", None)
|
|
347
|
+
if not thread_id:
|
|
348
|
+
ctx.dbg("Missing threadId")
|
|
349
|
+
return
|
|
350
|
+
user = context.get("user", None)
|
|
351
|
+
completed_at = g_db.get_thread_column(thread_id, "completedAt", user=user)
|
|
352
|
+
if completed_at:
|
|
353
|
+
context["completed"] = True
|
|
354
|
+
|
|
355
|
+
ctx.register_chat_tool_filter(tool_request)
|
|
356
|
+
|
|
357
|
+
async def chat_response(openai_response, context):
|
|
358
|
+
ctx.dbg("create_response")
|
|
359
|
+
o = openai_response
|
|
360
|
+
chat = context.get("chat")
|
|
361
|
+
usage = o.get("usage", None)
|
|
362
|
+
if not usage and not chat:
|
|
363
|
+
ctx.dbg("Missing chat and usage")
|
|
364
|
+
return
|
|
365
|
+
|
|
366
|
+
user = context.get("user", None)
|
|
367
|
+
thread_id = context.get("threadId", None)
|
|
368
|
+
provider = context.get("provider", None)
|
|
369
|
+
model_info = context.get("modelInfo", None)
|
|
370
|
+
model_cost = context.get("modelCost", model_info.get("cost", None)) or {"input": 0, "output": 0}
|
|
371
|
+
duration = context.get("duration", 0)
|
|
372
|
+
|
|
373
|
+
metadata = o.get("metadata", {})
|
|
374
|
+
choices = o.get("choices", [])
|
|
375
|
+
tasks = []
|
|
376
|
+
title = context.get("title") or prompt_to_title(ctx.last_user_prompt(chat) if chat else None)
|
|
377
|
+
completed_at = datetime.now()
|
|
378
|
+
|
|
379
|
+
model = model_info.get("name") or model_info.get("id")
|
|
380
|
+
finish_reason = choices[0].get("finish_reason", None) if len(choices) > 0 else None
|
|
381
|
+
input_price = model_cost.get("input", 0)
|
|
382
|
+
output_price = model_cost.get("output", 0)
|
|
383
|
+
input_tokens = usage.get("prompt_tokens", 0)
|
|
384
|
+
output_tokens = usage.get("completion_tokens", 0)
|
|
385
|
+
total_tokens = usage.get("total_tokens", input_tokens + output_tokens)
|
|
386
|
+
cost = o.get("cost", ((input_price * input_tokens) + (output_price * output_tokens)) / 1000000)
|
|
387
|
+
|
|
388
|
+
request = {
|
|
389
|
+
"user": user,
|
|
390
|
+
"model": model,
|
|
391
|
+
"duration": duration,
|
|
392
|
+
"cost": cost,
|
|
393
|
+
"inputPrice": input_price,
|
|
394
|
+
"inputTokens": input_tokens,
|
|
395
|
+
"inputCachedTokens": usage.get("inputCachedTokens", 0),
|
|
396
|
+
"outputPrice": output_price,
|
|
397
|
+
"outputTokens": output_tokens,
|
|
398
|
+
"finishReason": finish_reason,
|
|
399
|
+
"provider": provider,
|
|
400
|
+
"providerModel": o.get("model", None),
|
|
401
|
+
"providerRef": o.get("provider", None),
|
|
402
|
+
"threadId": thread_id,
|
|
403
|
+
"title": title,
|
|
404
|
+
"startedAt": context.get("startedAt"),
|
|
405
|
+
"totalTokens": total_tokens,
|
|
406
|
+
"usage": usage,
|
|
407
|
+
"completedAt": completed_at,
|
|
408
|
+
"toolHistory": o.get("tool_history", None),
|
|
409
|
+
"ref": o.get("id", None),
|
|
410
|
+
}
|
|
411
|
+
tasks.append(g_db.create_request_async(request, user=user))
|
|
412
|
+
|
|
413
|
+
if thread_id:
|
|
414
|
+
messages = chat.get("messages", [])
|
|
415
|
+
last_role = messages[-1].get("role", None) if len(messages) > 0 else None
|
|
416
|
+
if last_role == "user" or last_role == "tool":
|
|
417
|
+
user_message = messages[-1]
|
|
418
|
+
user_message["model"] = model
|
|
419
|
+
user_message["usage"] = {
|
|
420
|
+
"tokens": input_tokens,
|
|
421
|
+
"price": input_price,
|
|
422
|
+
"cost": (input_price * input_tokens) / 1000000,
|
|
423
|
+
}
|
|
424
|
+
else:
|
|
425
|
+
ctx.dbg(
|
|
426
|
+
f"Missing user message for thread {thread_id}, {len(messages)} messages, last role: {last_role}"
|
|
427
|
+
)
|
|
428
|
+
assistant_message = ctx.chat_response_to_message(o)
|
|
429
|
+
assistant_message["model"] = model
|
|
430
|
+
assistant_message["usage"] = {
|
|
431
|
+
"tokens": output_tokens,
|
|
432
|
+
"price": output_price,
|
|
433
|
+
"cost": (output_price * output_tokens) / 1000000,
|
|
434
|
+
"duration": duration,
|
|
435
|
+
}
|
|
436
|
+
messages.append(assistant_message)
|
|
437
|
+
|
|
438
|
+
update_thread = {
|
|
439
|
+
"model": model,
|
|
440
|
+
"providerModel": o.get("model"),
|
|
441
|
+
"modelInfo": model_info,
|
|
442
|
+
"messages": messages,
|
|
443
|
+
"completedAt": completed_at,
|
|
444
|
+
}
|
|
445
|
+
if "error" in metadata:
|
|
446
|
+
update_thread["error"] = metadata["error"]
|
|
447
|
+
tasks.append(g_db.update_thread_async(thread_id, update_thread, user=user))
|
|
448
|
+
else:
|
|
449
|
+
ctx.dbg("Missing thread_id")
|
|
450
|
+
|
|
451
|
+
await asyncio.gather(*tasks)
|
|
452
|
+
|
|
453
|
+
# Update thread costs from all thread requests
|
|
454
|
+
thread_requests = g_db.query_requests({"threadId": thread_id}, user=user)
|
|
455
|
+
total_costs = 0
|
|
456
|
+
total_input = 0
|
|
457
|
+
total_output = 0
|
|
458
|
+
for request in thread_requests:
|
|
459
|
+
total_costs += request.get("cost", 0) or 0
|
|
460
|
+
total_input += request.get("inputTokens", 0) or 0
|
|
461
|
+
total_output += request.get("outputTokens", 0) or 0
|
|
462
|
+
stats = {
|
|
463
|
+
"inputTokens": total_input,
|
|
464
|
+
"outputTokens": total_output,
|
|
465
|
+
"cost": total_costs,
|
|
466
|
+
"duration": duration,
|
|
467
|
+
"requests": len(thread_requests),
|
|
468
|
+
}
|
|
469
|
+
g_db.update_thread(
|
|
470
|
+
thread_id,
|
|
471
|
+
{
|
|
472
|
+
"inputTokens": total_input,
|
|
473
|
+
"outputTokens": total_output,
|
|
474
|
+
"cost": total_costs,
|
|
475
|
+
"stats": stats,
|
|
476
|
+
},
|
|
477
|
+
user=user,
|
|
478
|
+
)
|
|
479
|
+
|
|
480
|
+
ctx.register_chat_response_filter(chat_response)
|
|
481
|
+
|
|
482
|
+
async def chat_error(e: Exception, context: Any):
|
|
483
|
+
error = ctx.error_message(e)
|
|
484
|
+
ctx.dbg(f"Chat error: {error}")
|
|
485
|
+
chat = context.get("chat")
|
|
486
|
+
if not chat:
|
|
487
|
+
ctx.dbg("Missing chat")
|
|
488
|
+
return
|
|
489
|
+
|
|
490
|
+
title = context.get("title") or prompt_to_title(ctx.last_user_prompt(chat) if chat else None)
|
|
491
|
+
completed_at = datetime.now()
|
|
492
|
+
user = context.get("user", None)
|
|
493
|
+
|
|
494
|
+
thread_id = context.get("threadId", None)
|
|
495
|
+
tasks = []
|
|
496
|
+
if thread_id:
|
|
497
|
+
tasks.append(g_db.update_thread_async(thread_id, {"completedAt": completed_at, "error": error}, user=user))
|
|
498
|
+
else:
|
|
499
|
+
ctx.dbg("Missing threadId")
|
|
500
|
+
|
|
501
|
+
request = {
|
|
502
|
+
"user": user,
|
|
503
|
+
"model": chat.get("model", None),
|
|
504
|
+
"title": title,
|
|
505
|
+
"threadId": thread_id,
|
|
506
|
+
"startedAt": context.get("startedAt"),
|
|
507
|
+
"completedAt": completed_at,
|
|
508
|
+
"error": error,
|
|
509
|
+
"stackTrace": context.get("stackTrace", None),
|
|
510
|
+
}
|
|
511
|
+
tasks.append(g_db.create_request_async(request, user=user))
|
|
512
|
+
|
|
513
|
+
if len(tasks) > 0:
|
|
514
|
+
await asyncio.gather(*tasks)
|
|
515
|
+
|
|
516
|
+
ctx.register_chat_error_filter(chat_error)
|
|
517
|
+
|
|
518
|
+
|
|
519
|
+
__install__ = install
|
|
Binary file
|
|
Binary file
|
|
Binary file
|