dellweb3 commited on
Commit
52107a5
·
verified ·
1 Parent(s): d979b2d

Release 2026.11.12-hf-deepseek.1

Browse files

Published at 2026-11-12T17:52:36.880123+08:00. Minimal route fix; pending remote ERP acceptance. No credentials in source.

deploy/hf/app.py CHANGED
@@ -103,7 +103,18 @@ def providers_from_environment(root):
103
  def create_app():
104
  data=ROOT/'.local/hf-runtime';data.mkdir(parents=True,exist_ok=True)
105
  providers=providers_from_environment(data)
 
 
 
 
 
106
  engine=LocalWorkbench(data,providers,sandbox_capacity=int(os.environ.get('ZKCODING_SANDBOX_CAPACITY','3')))
107
  engine.identity.seed_demo_users();engine.preview_origin=ORIGIN
108
  threading.Thread(target=initialize_model_health,args=(providers,data/'model-health.json'),daemon=True,name='space-model-health').start()
109
- return SpaceIngress(create_web(engine,7860),ORIGIN,bridge=PreviewBridge(engine))
 
 
 
 
 
 
 
103
  def create_app():
104
  data=ROOT/'.local/hf-runtime';data.mkdir(parents=True,exist_ok=True)
105
  providers=providers_from_environment(data)
106
+ preferred=os.environ.get('ZKCODING_DEFAULT_MODEL','')
107
+ if preferred:
108
+ if not any(p.model==preferred and p.selectable for p in providers):
109
+ raise ValueError('INVALID_DEFAULT_MODEL')
110
+ for provider in providers:provider.automatic_preference=preferred
111
  engine=LocalWorkbench(data,providers,sandbox_capacity=int(os.environ.get('ZKCODING_SANDBOX_CAPACITY','3')))
112
  engine.identity.seed_demo_users();engine.preview_origin=ORIGIN
113
  threading.Thread(target=initialize_model_health,args=(providers,data/'model-health.json'),daemon=True,name='space-model-health').start()
114
+ web=create_web(engine,7860)
115
+ release_path=ROOT/'deploy/hf/release.json'
116
+ if release_path.is_file():
117
+ release=json.loads(release_path.read_text(encoding='utf-8'))
118
+ @web.get('/api/release')
119
+ def release_info():return release
120
+ return SpaceIngress(web,ORIGIN,bridge=PreviewBridge(engine))
deploy/hf/release.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "version": "2026.11.12-hf-deepseek.1",
3
+ "published_at": "2026-11-12T17:52:36.880123+08:00",
4
+ "space": "dellweb3/VibeCoding",
5
+ "base_hf_commit": "d979b2df9f6dcb96166806901bfacc6268e58b83",
6
+ "purpose": "DeepSeek 4.1 default route and configured Bailian credential refresh",
7
+ "files_sha256": {
8
+ "deploy/hf/app.py": "e401c97914fa5c51043c9e002e0b3ba176d999fa235150f7626cefe5042c94a9",
9
+ "src/modules/llm_gateway/routing.py": "21c9c617121d50e0395adcd8b95b0a60ab1fbe3f866ae7d38b73f6ea81d9d0d0"
10
+ }
11
+ }
src/modules/llm_gateway/routing.py CHANGED
@@ -10,16 +10,14 @@ PROFILES = {
10
  'qwen3.8-max': {'roles':['design','diagnose','code','repair'], 'json_mode':True, 'thinking':False},
11
  }
12
 
13
- # Quota burn-down order per task kind: spend soonest-expiring quota first and keep the
14
- # longest-dated model as reserve. Models missing from a list keep configured order after it.
15
- # 2026-10-28 quota evidence: deepseek-v4-flash-0731 expires 10/31, kimi-k3 11/18,
16
- # qwen3.8-max-0902 12/01, deepseek-v4.1-flash-1017 01/17.
17
  PREFERENCE = {
18
- 'chat': ['deepseek-v4-flash-0731', 'kimi-k3', 'qwen3.8-max-0902', 'deepseek-v4.1-flash-1017'],
19
- 'design': ['kimi-k3', 'deepseek-v4-flash-0731', 'qwen3.8-max-0902', 'deepseek-v4.1-flash-1017'],
20
- 'diagnose': ['kimi-k3', 'deepseek-v4-flash-0731', 'qwen3.8-max-0902', 'deepseek-v4.1-flash-1017'],
21
- 'code': ['qwen3.8-max-0902', 'deepseek-v4-flash-0731', 'kimi-k3', 'deepseek-v4.1-flash-1017'],
22
- 'repair': ['qwen3.8-max-0902', 'deepseek-v4-flash-0731', 'kimi-k3', 'deepseek-v4.1-flash-1017'],
23
  }
24
 
25
 
@@ -55,6 +53,10 @@ def plan(providers, kind, run):
55
  order=PREFERENCE.get(kind,[])
56
  if kind in {'chat','design','diagnose','code','repair'}:
57
  order=['gpt-6.1-sol','gpt-6-astra','gpt-5.5',*order]
 
 
 
 
58
  available.sort(key=lambda p: order.index(p.model) if p.model in order else len(order))
59
  return available,('openai-medium-priority' if available and getattr(available[0],'provider_family',None)=='openai' else 'configured-verified-order')
60
 
 
10
  'qwen3.8-max': {'roles':['design','diagnose','code','repair'], 'json_mode':True, 'thinking':False},
11
  }
12
 
13
+ # 2026-11-12: user requests DeepSeek 4.1 when Sol is unavailable.
14
+ # Models still require successful health checks; manual selections remain pinned.
 
 
15
  PREFERENCE = {
16
+ 'chat': ['deepseek-v4.1-flash-1017', 'deepseek-v4-flash-0731', 'kimi-k3', 'qwen3.8-max-0902'],
17
+ 'design': ['deepseek-v4.1-flash-1017', 'kimi-k3', 'deepseek-v4-flash-0731', 'qwen3.8-max-0902'],
18
+ 'diagnose': ['deepseek-v4.1-flash-1017', 'kimi-k3', 'deepseek-v4-flash-0731', 'qwen3.8-max-0902'],
19
+ 'code': ['deepseek-v4.1-flash-1017', 'qwen3.8-max-0902', 'deepseek-v4-flash-0731', 'kimi-k3'],
20
+ 'repair': ['deepseek-v4.1-flash-1017', 'qwen3.8-max-0902', 'deepseek-v4-flash-0731', 'kimi-k3'],
21
  }
22
 
23
 
 
53
  order=PREFERENCE.get(kind,[])
54
  if kind in {'chat','design','diagnose','code','repair'}:
55
  order=['gpt-6.1-sol','gpt-6-astra','gpt-5.5',*order]
56
+ preferred={getattr(p,'automatic_preference',None) for p in providers}-{None}
57
+ if len(preferred)==1:
58
+ first=preferred.pop()
59
+ order=[first,*[model for model in order if model!=first]]
60
  available.sort(key=lambda p: order.index(p.model) if p.model in order else len(order))
61
  return available,('openai-medium-priority' if available and getattr(available[0],'provider_family',None)=='openai' else 'configured-verified-order')
62