Update configuration paths and enhance parameter options in the model configuration

- Changed configuration file paths from "configs" to "config" in CUDA and LMDeploy installers.
- Added new parameters with options for tensor split and main GPU in the parameter registry.
- Enhanced the ModelConfig view to support dropdowns for parameters with options, improving user experience.
- Updated styles for dropdown components to align with the new design specifications.

Author: lapy

Dockerfile

@@ -167,7 +167,7 @@ RUN ln -sf /usr/bin/python3 /usr/bin/python
 
 # Create non-root user and data directory structure
 RUN useradd -m -s /bin/bash appuser && \
-    mkdir -p /app/data/models /app/data/config /app/data/configs /app/data/logs /app/data/llama-cpp /app/data/hf-cache/hub && \
+    mkdir -p /app/data/models /app/data/config /app/data/logs /app/data/llama-cpp /app/data/hf-cache/hub && \
     chown -R appuser:appuser /app && \
     # Ensure entrypoint script is accessible to appuser
     chmod 755 /usr/local/bin/docker-entrypoint.sh

backend/cuda_installer.py

@@ -100,7 +100,7 @@ def __init__(
 
         log_path = log_path or os.path.join(data_root, "logs", "cuda_install.log")
         state_path = state_path or os.path.join(
-            data_root, "configs", "cuda_installer.json"
+            data_root, "config", "cuda_installer.json"
         )
         download_dir = download_dir or os.path.join(
             data_root, "temp", "cuda_installers"

backend/lmdeploy_manager.py

@@ -63,7 +63,7 @@ def __init__(
     self._venv_path = os.path.join(self._base_dir, "venv")
     log_path = log_path or os.path.join(data_root, "logs", "lmdeploy_install.log")
     state_path = state_path or os.path.join(
-      data_root, "configs", "lmdeploy_manager.json"
+      data_root, "config", "lmdeploy_manager.json"
     )
     self._log_path = os.path.abspath(log_path)
     self._state_path = os.path.abspath(state_path)

backend/param_registry.py

@@ -6,7 +6,8 @@
 import copy
 from typing import Any, Dict, List
 
-# Param entry: key, label, type ("int"|"float"|"bool"|"string"), default, min, max (optional), description (optional)
+# Param entry: key, label, type ("int"|"float"|"bool"|"string"), default, min, max (optional), description (optional), options (optional)
+# options: list of {"value": ..., "label": "..."} for select/dropdown; when present, frontend renders a select instead of text/number input
 ParamDef = Dict[str, Any]
 
 # Basic params shown by default (most common for chat/embedding)
@@ -15,51 +16,51 @@
     {"key": "model_alias", "label": "Model alias", "type": "string", "default": "", "description": "Expose this model under a custom runtime ID instead of the default Hugging Face-derived name"},
     {"key": "ctx_size", "label": "Context size", "type": "int", "default": 2048, "min": 512, "max": 1_000_000, "description": "Maximum context length in tokens"},
     {"key": "n_gpu_layers", "label": "GPU layers", "type": "int", "default": -1, "min": -1, "max": 1000, "description": "Number of layers to offload to GPU (-1 = all)"},
+    {"key": "tensor_split", "label": "Tensor split", "type": "string", "default": "", "description": "Comma-separated GPU allocation (e.g. 40,30,30)"},
+    {"key": "main_gpu", "label": "Main GPU", "type": "int", "default": 0, "min": 0, "description": "Index of the main GPU"},
     {"key": "batch_size", "label": "Batch size", "type": "int", "default": 512, "min": 1, "max": 2048, "description": "Batch size for prompt processing"},
     {"key": "threads", "label": "Threads", "type": "int", "default": 4, "min": 1, "max": 64, "description": "Number of threads"},
     {"key": "embedding", "label": "Embedding mode", "type": "bool", "default": False, "description": "Enable embedding-only mode"},
 ]
 
 # Advanced params (shown in expandable "Advanced" section)
 LLAMA_CPP_ADVANCED: List[ParamDef] = [
-    {"key": "n_predict", "label": "Max tokens to predict", "type": "int", "default": -1, "min": -1, "max": 100_000},
-    {"key": "ubatch_size", "label": "Ubatch size", "type": "int", "default": 512, "min": 1, "max": 2048},
-    {"key": "temp", "label": "Temperature", "type": "float", "default": 0.8, "min": 0, "max": 2},
-    {"key": "top_k", "label": "Top K", "type": "int", "default": 40, "min": 0, "max": 1000},
-    {"key": "top_p", "label": "Top P", "type": "float", "default": 0.9, "min": 0, "max": 1},
-    {"key": "min_p", "label": "Min P", "type": "float", "default": 0.0, "min": 0, "max": 1},
-    {"key": "typical_p", "label": "Typical P", "type": "float", "default": 1.0, "min": 0, "max": 1},
-    {"key": "repeat_penalty", "label": "Repeat penalty", "type": "float", "default": 1.1, "min": 1, "max": 2},
-    {"key": "presence_penalty", "label": "Presence penalty", "type": "float", "default": 0, "min": -2, "max": 2},
-    {"key": "frequency_penalty", "label": "Frequency penalty", "type": "float", "default": 0, "min": -2, "max": 2},
-    {"key": "seed", "label": "Seed", "type": "int", "default": -1, "min": -1, "max": 2**31 - 1},
-    {"key": "threads_batch", "label": "Threads (batch)", "type": "int", "default": -1, "min": -1, "max": 64},
-    {"key": "parallel", "label": "Parallel", "type": "int", "default": 1, "min": 1, "max": 64},
-    {"key": "rope_freq_base", "label": "RoPE freq base", "type": "float", "default": 0, "min": 0},
-    {"key": "rope_freq_scale", "label": "RoPE freq scale", "type": "float", "default": 0, "min": 0},
-    {"key": "flash_attn", "label": "Flash attention", "type": "bool", "default": False},
-    {"key": "yarn_ext_factor", "label": "YaRN ext factor", "type": "float", "default": -1, "min": -1},
-    {"key": "yarn_attn_factor", "label": "YaRN attn factor", "type": "float", "default": 1, "min": 0},
-    {"key": "no_mmap", "label": "No mmap", "type": "bool", "default": False},
-    {"key": "mlock", "label": "MLock", "type": "bool", "default": False},
-    {"key": "low_vram", "label": "Low VRAM", "type": "bool", "default": False},
-    {"key": "logits_all", "label": "Logits all", "type": "bool", "default": False},
-    {"key": "cont_batching", "label": "Continuous batching", "type": "bool", "default": True},
-    {"key": "no_kv_offload", "label": "No KV offload", "type": "bool", "default": False},
-    {"key": "tensor_split", "label": "Tensor split", "type": "string", "default": ""},
-    {"key": "main_gpu", "label": "Main GPU", "type": "int", "default": 0, "min": 0},
-    {"key": "split_mode", "label": "Split mode", "type": "string", "default": ""},
-    {"key": "cache_type_k", "label": "Cache type K", "type": "string", "default": ""},
-    {"key": "cache_type_v", "label": "Cache type V", "type": "string", "default": ""},
-    {"key": "grammar", "label": "Grammar", "type": "string", "default": ""},
-    {"key": "json_schema", "label": "JSON schema", "type": "string", "default": ""},
-    {"key": "cpu_moe", "label": "CPU MoE", "type": "bool", "default": False},
-    {"key": "n_cpu_moe", "label": "N CPU MoE", "type": "int", "default": 0, "min": 0},
-    {"key": "override_tensor", "label": "Override tensor", "type": "string", "default": ""},
-    {"key": "rope_scaling", "label": "RoPE scaling", "type": "string", "default": ""},
-    {"key": "mirostat", "label": "Mirostat", "type": "int", "default": 0, "min": 0, "max": 2},
-    {"key": "mirostat_tau", "label": "Mirostat tau", "type": "float", "default": 5.0, "min": 0},
-    {"key": "mirostat_eta", "label": "Mirostat eta", "type": "float", "default": 0.1, "min": 0},
+    {"key": "n_predict", "label": "Max tokens to predict", "type": "int", "default": -1, "min": -1, "max": 100_000, "description": "Maximum tokens to generate (-1 = no limit)"},
+    {"key": "ubatch_size", "label": "Ubatch size", "type": "int", "default": 512, "min": 1, "max": 2048, "description": "Micro-batch size for prompt processing"},
+    {"key": "temp", "label": "Temperature", "type": "float", "default": 0.8, "min": 0, "max": 2, "description": "Sampling temperature (higher = more random)"},
+    {"key": "top_k", "label": "Top K", "type": "int", "default": 40, "min": 0, "max": 1000, "description": "Limit sampling to top K tokens (0 = disabled)"},
+    {"key": "top_p", "label": "Top P", "type": "float", "default": 0.9, "min": 0, "max": 1, "description": "Nucleus sampling: keep smallest set with cumulative probability"},
+    {"key": "min_p", "label": "Min P", "type": "float", "default": 0.0, "min": 0, "max": 1, "description": "Minimum token probability for sampling"},
+    {"key": "typical_p", "label": "Typical P", "type": "float", "default": 1.0, "min": 0, "max": 1, "description": "Typical sampling (1 = disabled)"},
+    {"key": "repeat_penalty", "label": "Repeat penalty", "type": "float", "default": 1.1, "min": 1, "max": 2, "description": "Penalize repeated tokens"},
+    {"key": "presence_penalty", "label": "Presence penalty", "type": "float", "default": 0, "min": -2, "max": 2, "description": "Penalize tokens that have appeared"},
+    {"key": "frequency_penalty", "label": "Frequency penalty", "type": "float", "default": 0, "min": -2, "max": 2, "description": "Penalize by token frequency"},
+    {"key": "seed", "label": "Seed", "type": "int", "default": -1, "min": -1, "max": 2**31 - 1, "description": "Random seed for reproducibility (-1 = random)"},
+    {"key": "threads_batch", "label": "Threads (batch)", "type": "int", "default": -1, "min": -1, "max": 64, "description": "Threads for batch inference (-1 = use threads)"},
+    {"key": "parallel", "label": "Parallel", "type": "int", "default": 1, "min": 1, "max": 64, "description": "Number of parallel sequences"},
+    {"key": "rope_freq_base", "label": "RoPE freq base", "type": "float", "default": 0, "min": 0, "description": "RoPE base frequency (0 = from model)"},
+    {"key": "rope_freq_scale", "label": "RoPE freq scale", "type": "float", "default": 0, "min": 0, "description": "RoPE frequency scaling (0 = from model)"},
+    {"key": "flash_attn", "label": "Flash attention", "type": "bool", "default": False, "description": "Use Flash Attention when available"},
+    {"key": "yarn_ext_factor", "label": "YaRN ext factor", "type": "float", "default": -1, "min": -1, "description": "YaRN extension factor (-1 = disabled)"},
+    {"key": "yarn_attn_factor", "label": "YaRN attn factor", "type": "float", "default": 1, "min": 0, "description": "YaRN attention factor"},
+    {"key": "no_mmap", "label": "No mmap", "type": "bool", "default": False, "description": "Disable memory-mapped model loading"},
+    {"key": "mlock", "label": "MLock", "type": "bool", "default": False, "description": "Lock model in RAM to prevent swapping"},
+    {"key": "low_vram", "label": "Low VRAM", "type": "bool", "default": False, "description": "Reduce VRAM use (slower inference)"},
+    {"key": "logits_all", "label": "Logits all", "type": "bool", "default": False, "description": "Return logits for all tokens (embedding mode)"},
+    {"key": "cont_batching", "label": "Continuous batching", "type": "bool", "default": True, "description": "Use continuous batching for higher throughput"},
+    {"key": "no_kv_offload", "label": "No KV offload", "type": "bool", "default": False, "description": "Disable KV cache offload to CPU"},
+    {"key": "split_mode", "label": "Split mode", "type": "string", "default": "", "description": "Tensor split mode (e.g. layer, row)", "options": [{"value": "", "label": "Default"}, {"value": "layer", "label": "Layer"}, {"value": "row", "label": "Row"}]},
+    {"key": "cache_type_k", "label": "Cache type K", "type": "string", "default": "", "description": "K cache backend (e.g. f16, q8_0)", "options": [{"value": "", "label": "Default"}, {"value": "f16", "label": "f16"}, {"value": "q8_0", "label": "q8_0"}, {"value": "q4_0", "label": "q4_0"}, {"value": "q4_1", "label": "q4_1"}]},
+    {"key": "cache_type_v", "label": "Cache type V", "type": "string", "default": "", "description": "V cache backend (e.g. f16, q8_0)", "options": [{"value": "", "label": "Default"}, {"value": "f16", "label": "f16"}, {"value": "q8_0", "label": "q8_0"}, {"value": "q4_0", "label": "q4_0"}, {"value": "q4_1", "label": "q4_1"}]},
+    {"key": "grammar", "label": "Grammar", "type": "string", "default": "", "description": "BNF grammar for constrained output"},
+    {"key": "json_schema", "label": "JSON schema", "type": "string", "default": "", "description": "JSON schema for structured output"},
+    {"key": "cpu_moe", "label": "CPU MoE", "type": "bool", "default": False, "description": "Run MoE experts on CPU"},
+    {"key": "n_cpu_moe", "label": "N CPU MoE", "type": "int", "default": 0, "min": 0, "description": "Number of MoE experts to run on CPU"},
+    {"key": "override_tensor", "label": "Override tensor", "type": "string", "default": "", "description": "Override tensor layout (e.g. for debugging)"},
+    {"key": "rope_scaling", "label": "RoPE scaling", "type": "string", "default": "", "description": "RoPE scaling type (e.g. yarn, linear)", "options": [{"value": "", "label": "Default"}, {"value": "yarn", "label": "YaRN"}, {"value": "linear", "label": "Linear"}, {"value": "none", "label": "None"}]},
+    {"key": "mirostat", "label": "Mirostat", "type": "int", "default": 0, "min": 0, "max": 2, "description": "Mirostat mode (0=off, 1=mirostat, 2=mirostat 2.0)", "options": [{"value": 0, "label": "Off"}, {"value": 1, "label": "Mirostat 1"}, {"value": 2, "label": "Mirostat 2.0"}]},
+    {"key": "mirostat_tau", "label": "Mirostat tau", "type": "float", "default": 5.0, "min": 0, "description": "Mirostat target perplexity"},
+    {"key": "mirostat_eta", "label": "Mirostat eta", "type": "float", "default": 0.1, "min": 0, "description": "Mirostat learning rate"},
 ]
 
 # ik_llama.cpp: same as llama_cpp plus these extras (and different mirostat flag names)
@@ -78,8 +79,8 @@
     {"key": "tensor_parallel", "label": "Tensor parallel", "type": "int", "default": 1, "min": 1, "max": 8, "description": "Tensor parallelism degree"},
 ]
 LMDEPLOY_ADVANCED: List[ParamDef] = [
-    {"key": "dtype", "label": "Dtype", "type": "string", "default": "auto", "description": "Model dtype (auto, float16, bfloat16)"},
-    {"key": "quant_policy", "label": "Quantization policy", "type": "int", "default": 0, "min": 0, "max": 8, "description": "KV cache quantization (0=off, 4=4bit, 8=8bit)"},
+    {"key": "dtype", "label": "Dtype", "type": "string", "default": "auto", "description": "Model dtype (auto, float16, bfloat16)", "options": [{"value": "auto", "label": "Auto"}, {"value": "float16", "label": "float16"}, {"value": "bfloat16", "label": "bfloat16"}]},
+    {"key": "quant_policy", "label": "Quantization policy", "type": "int", "default": 0, "min": 0, "max": 8, "description": "KV cache quantization (0=off, 4=4bit, 8=8bit)", "options": [{"value": 0, "label": "Off"}, {"value": 4, "label": "4-bit"}, {"value": 8, "label": "8-bit"}]},
     {"key": "enable_prefix_caching", "label": "Prefix caching", "type": "bool", "default": False, "description": "Enable prefix caching"},
     {"key": "chat_template", "label": "Chat template", "type": "string", "default": "", "description": "Override chat template"},
 ]

frontend/src/styles/_components.css

@@ -605,16 +605,98 @@
   box-shadow: var(--shadow-lg);
 }
 
+/* Dropdown header and filter: no white border, theme colors only */
+.p-dropdown-panel .p-dropdown-header {
+  padding: 0.5rem 0.75rem;
+  border: none;
+  outline: none;
+  box-shadow: none;
+  background: transparent;
+  color: var(--text-secondary);
+}
+.p-dropdown-panel .p-dropdown-header * {
+  border-color: var(--border-primary);
+}
+.p-dropdown-panel .p-dropdown-filter-container {
+  display: flex;
+  align-items: center;
+  height: 2.375rem;
+  border: 1px solid var(--border-primary);
+  border-radius: var(--radius-md);
+  background: var(--bg-surface);
+  color: var(--text-primary);
+  box-sizing: border-box;
+  position: relative;
+}
+/* Fallback search icon (visible when PrimeVue SVG is not) */
+.p-dropdown-panel .p-dropdown-filter-container::before {
+  content: '';
+  flex-shrink: 0;
+  width: 1rem;
+  height: 1rem;
+  margin: 0 0.5rem;
+  background-color: var(--text-secondary);
+  mask-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 24 24' fill='none' stroke='black' stroke-width='2' stroke-linecap='round' stroke-linejoin='round'%3E%3Ccircle cx='11' cy='11' r='8'/%3E%3Cpath d='m21 21-4.35-4.35'/%3E%3C/svg%3E");
+  mask-size: contain;
+  mask-repeat: no-repeat;
+  mask-position: center;
+  -webkit-mask-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 24 24' fill='none' stroke='black' stroke-width='2' stroke-linecap='round' stroke-linejoin='round'%3E%3Ccircle cx='11' cy='11' r='8'/%3E%3Cpath d='m21 21-4.35-4.35'/%3E%3C/svg%3E");
+  -webkit-mask-size: contain;
+  -webkit-mask-repeat: no-repeat;
+  -webkit-mask-position: center;
+}
+.p-dropdown-panel .p-dropdown-filter-container:focus-within {
+  outline: none;
+  border-color: var(--accent-cyan);
+  box-shadow: 0 0 0 3px var(--focus-ring);
+}
+/* Hide built-in filter icon; we use ::before on container as fallback */
+.p-dropdown-panel .p-dropdown-filter-container .p-dropdown-filter-icon,
+.p-dropdown-panel .p-dropdown-filter-container [data-pc-section="filtericon"] {
+  display: none !important;
+}
+.p-dropdown-panel .p-dropdown-filter {
+  flex: 1;
+  min-width: 0;
+  height: 100%;
+  padding: 0 0.5rem;
+  border: none;
+  background: transparent;
+  color: var(--text-primary);
+  border-radius: 0;
+  box-shadow: none;
+}
+.p-dropdown-panel .p-dropdown-filter::placeholder {
+  color: var(--text-muted);
+}
+
 .p-dropdown-panel .p-dropdown-items .p-dropdown-item {
   color: var(--text-primary);
   background: transparent;
+  padding-left: 0.75rem;
+  padding-right: 0.75rem;
 }
 
 .p-dropdown-panel .p-dropdown-items .p-dropdown-item:hover {
   background: var(--hover-bg);
 }
 
+/* InputNumber: fixed height and contain inner input (no overflow) */
+.p-inputnumber {
+  height: 2.375rem;
+  min-height: 2.375rem;
+  padding: 0 0.25rem;
+  display: flex;
+  align-items: center;
+  box-sizing: border-box;
+}
+
 .p-inputnumber .p-inputnumber-input {
+  height: 2rem;
+  max-height: 100%;
+  padding: 0 0.25rem;
+  line-height: 2rem;
+  box-sizing: border-box;
   background: transparent;
   color: var(--text-primary);
   border: none;