fix: add nv_framework back for API safety

- Remove /bin/bash -ic wrapper from sut_config.command
- Use list format for all command arguments (matches platform auto-gen)
- Add nv_framework field
- Remove unnecessary fields: docker_image, nv_docker_image, modelFormat, model, cpu_num
- Fix ref_config to match platform format (port 80, max_model_len 4096)

main.py

@@ -121,12 +121,12 @@ def init_db():
 # ModelScope 搜索
 # ============================================================
 
-def search_models(keyword: str, limit: int = 100) -> list:
+def search_models(keyword: str, limit: int = 50) -> list:
     """从 ModelScope 搜索模型"""
     url = "https://modelscope.cn/openapi/v1/models"
     params = {
         'search': keyword,
-        'page_size': limit,
+        'page_size': min(limit, 50),  # API 上限 50
         'page_number': 1,
         'sort': 'downloads',
     }
@@ -135,7 +135,11 @@ def search_models(keyword: str, limit: int = 100) -> list:
                             headers={'User-Agent': 'Mozilla/5.0'})
         data = resp.json()
         if data.get('success'):
-            return data.get('data', {}).get('models', [])
+            models = data.get('data', {}).get('models', [])
+            log(f"  搜索 [{keyword}]: status={resp.status_code} models={len(models)}")
+            return models
+        else:
+            log(f"  搜索 [{keyword}]: success=false, data={str(data)[:200]}")
     except Exception as e:
         log(f"  搜索失败 [{keyword}]: {e}")
     return []
@@ -191,7 +195,8 @@ def check_platform_verify(model_id: str) -> dict:
         resp = requests.get(url, headers=headers, params={'modelId': model_id}, timeout=10)
         data = resp.json()
         if data.get('code') == 0:
-            return data.get('data', {}).get('verifyResult', {})
+            result = data.get('data', {}).get('verifyResult', {})
+            return result if isinstance(result, dict) else {}
     except Exception:
         pass
     return {}
@@ -238,16 +243,15 @@ def check_queue_available(gpu: str, token: str) -> int:
 
 
 def build_config_params(gpu: str) -> str:
-    """构建 YAML 配置"""
+    """构建 YAML 配置 - 完全匹配平台自动生成的格式"""
     config = GPU_CONFIGS.get(gpu, {})
     framework = config.get('framework', 'vllm')
-    docker_image = config.get('docker_image', '')
 
     if framework == 'llama.cpp':
+        # hygon_k100-ai 使用 llama.cpp
         params = {
             'framework': 'llama.cpp',
-            'docker_image': docker_image,
+            'nv_framework': 'llama.cpp',
-            'nv_docker_image': docker_image,
             'api': 'completion',
             'max_tokens': 1024,
             'temperature': 0.7,
@@ -255,21 +259,21 @@ def build_config_params(gpu: str) -> str:
             'top_p': 0.9,
             'lang': 'zh',
             'max_model_len': 4096,
-            'modelFormat': 'GGUF',
             'sut_config': {
-                'values': {
                 'gpu_num': 1,
+                'values': {
                     'command': [
-                        '/bin/bash', '-ic',
+                        'llama-server', '--model', '/model', '--alias', 'llm',
-                        'llama-server --model /model --alias llm --threads 20 '
+                        '--threads', '20', '--n-gpu-layers', '999', '--prio', '3',
-                        '--n-gpu-layers 999 --prio 3 --min_p 0.01 '
+                        '--min_p', '0.01', '--ctx-size', '4096',
-                        '--ctx-size 4096 --host 0.0.0.0 --port 8000 --jinja --flash-attn off'
+                        '--host', '0.0.0.0', '--port', '8000',
+                        '--jinja', '--flash-attn', 'off',
                     ]
                 }
             },
             'ref_config': {
+                'gpu_num': 1,
                 'values': {
-                    'cpu_num': 2, 'gpu_num': 1,
                     'command': [
                         'llama-server', '--model', '/model', '--alias', 'llm',
                         '--threads', '20', '--n-gpu-layers', '999',
@@ -277,13 +281,12 @@ def build_config_params(gpu: str) -> str:
                     ]
                 }
             },
-            'model': 'llm',
         }
     else:
+        # vllm (P800, Ascend, MetaX)
         params = {
             'framework': 'vllm',
-            'docker_image': docker_image,
+            'nv_framework': 'vllm',
-            'nv_docker_image': docker_image,
             'api': 'completion',
             'max_tokens': 1024,
             'temperature': 0.7,
@@ -291,21 +294,9 @@ def build_config_params(gpu: str) -> str:
             'top_p': 0.9,
             'lang': 'zh',
             'max_model_len': 2048,
-            'modelFormat': 'HuggingFace',
             'sut_config': {
-                'values': {
                 'gpu_num': 1,
-                    'command': [
-                        '/bin/bash', '-ic',
-                        'vllm serve /model --port 8000 --served-model-name llm '
-                        '--max-model-len 2048 --dtype auto --gpu-memory-utilization 0.95 '
-                        '-tp 1 --enforce-eager --trust-remote-code'
-                    ]
-                }
-            },
-            'ref_config': {
                 'values': {
-                    'cpu_num': 2, 'gpu_num': 1,
                     'command': [
                         'vllm', 'serve', '/model', '--port', '8000',
                         '--served-model-name', 'llm', '--max-model-len', '2048',
@@ -314,9 +305,18 @@ def build_config_params(gpu: str) -> str:
                     ]
                 }
             },
-            'model': 'llm',
+            'ref_config': {
+                'gpu_num': 1,
+                'values': {
+                    'command': [
+                        'vllm', 'serve', '/model', '--port', '80',
+                        '--served-model-name', 'llm', '--max-model-len', '4096',
+                        '--enforce-eager', '--trust-remote-code', '-tp', '1',
+                    ]
                 }
-    return yaml.dump(params, default_flow_style=False, allow_unicode=True)
+            },
+        }
+    return yaml.dump(params, default_flow_style=False, allow_unicode=True, width=1000)
 
 
 def submit_model(model_url: str, gpu: str, token: str) -> tuple: