feat: v1.4.0 - 五项优化
1. 模型选择: 点击搜索框弹出top5模型，支持搜索过滤 2. 选模型后自动调整参数: ctx_size设为模型默认上下文，n_gpu_layers设为all - models表新增 default_ctx 字段，各模型预设上下文值 3. 参数配置新增「已改参数」标签页 - 放在最右边，绿色样式区别于其他标签 - 显示所有被修改过(非默认值)的参数 4. 复制按钮移到命令文本框左上角悬空 5. 自然语言默认提示文本可在后台管理中编辑 - 新增 /api/settings/public 接口 - 后台系统设置新增 nl_default_text 配置项
This commit is contained in:
@@ -88,6 +88,7 @@ def init_db():
|
||||
kv_heads INTEGER NOT NULL,
|
||||
head_dim INTEGER NOT NULL,
|
||||
attention_heads INTEGER NOT NULL,
|
||||
default_ctx INTEGER DEFAULT 0,
|
||||
quant TEXT DEFAULT '',
|
||||
description TEXT DEFAULT '',
|
||||
sort_order INTEGER DEFAULT 0
|
||||
@@ -313,66 +314,66 @@ def insert_default_data(conn):
|
||||
(v2_id,) + p)
|
||||
|
||||
# ===== Default models =====
|
||||
# (base_model, name, size_gb, layers, embd, kv_heads, head_dim, attention_heads, quant, description, sort_order)
|
||||
# (base_model, name, size_gb, layers, embd, kv_heads, head_dim, attention_heads, default_ctx, quant, description, sort_order)
|
||||
default_models = [
|
||||
# Llama-3-8B-Instruct
|
||||
("Llama-3-8B-Instruct", "Llama-3-8B-Instruct (Q4_K_M)", 4.9, 32, 4096, 8, 128, 32, "Q4_K_M", "Meta Llama 3 8B Instruct, Q4_K_M 量化", 1),
|
||||
("Llama-3-8B-Instruct", "Llama-3-8B-Instruct (Q8_0)", 8.5, 32, 4096, 8, 128, 32, "Q8_0", "Meta Llama 3 8B Instruct, Q8_0 量化", 2),
|
||||
("Llama-3-8B-Instruct", "Llama-3-8B-Instruct (FP16)", 15.5, 32, 4096, 8, 128, 32, "FP16", "Meta Llama 3 8B Instruct, FP16", 3),
|
||||
# Llama-3-70B-Instruct
|
||||
("Llama-3-70B-Instruct", "Llama-3-70B-Instruct (Q4_K_M)", 38.5, 80, 8192, 8, 128, 64, "Q4_K_M", "Meta Llama 3 70B Instruct, Q4_K_M 量化", 4),
|
||||
("Llama-3-70B-Instruct", "Llama-3-70B-Instruct (Q8_0)", 74.0, 80, 8192, 8, 128, 64, "Q8_0", "Meta Llama 3 70B Instruct, Q8_0 量化", 5),
|
||||
("Llama-3-70B-Instruct", "Llama-3-70B-Instruct (FP16)", 138.0, 80, 8192, 8, 128, 64, "FP16", "Meta Llama 3 70B Instruct, FP16", 6),
|
||||
# Llama-3.1-8B-Instruct
|
||||
("Llama-3.1-8B-Instruct", "Llama-3.1-8B-Instruct (Q4_K_M)", 4.9, 32, 4096, 8, 128, 32, "Q4_K_M", "Meta Llama 3.1 8B Instruct, Q4_K_M 量化", 7),
|
||||
("Llama-3.1-8B-Instruct", "Llama-3.1-8B-Instruct (Q8_0)", 8.5, 32, 4096, 8, 128, 32, "Q8_0", "Meta Llama 3.1 8B Instruct, Q8_0 量化", 8),
|
||||
# Llama-3.1-70B-Instruct
|
||||
("Llama-3.1-70B-Instruct", "Llama-3.1-70B-Instruct (Q4_K_M)", 38.5, 80, 8192, 8, 128, 64, "Q4_K_M", "Meta Llama 3.1 70B Instruct, Q4_K_M 量化", 9),
|
||||
("Llama-3.1-70B-Instruct", "Llama-3.1-70B-Instruct (Q8_0)", 74.0, 80, 8192, 8, 128, 64, "Q8_0", "Meta Llama 3.1 70B Instruct, Q8_0 量化", 10),
|
||||
# Llama-3.1-405B-Instruct
|
||||
("Llama-3.1-405B-Instruct", "Llama-3.1-405B-Instruct (Q4_K_M)", 226.0, 126, 16384, 8, 128, 128, "Q4_K_M", "Meta Llama 3.1 405B Instruct, Q4_K_M 量化", 11),
|
||||
# Qwen2.5-7B-Instruct
|
||||
("Qwen2.5-7B-Instruct", "Qwen2.5-7B-Instruct (Q4_K_M)", 4.7, 28, 3584, 4, 128, 28, "Q4_K_M", "Qwen2.5 7B Instruct, Q4_K_M 量化", 12),
|
||||
("Qwen2.5-7B-Instruct", "Qwen2.5-7B-Instruct (Q8_0)", 7.6, 28, 3584, 4, 128, 28, "Q8_0", "Qwen2.5 7B Instruct, Q8_0 量化", 13),
|
||||
# Qwen2.5-14B-Instruct
|
||||
("Qwen2.5-14B-Instruct", "Qwen2.5-14B-Instruct (Q4_K_M)", 8.7, 40, 5120, 8, 128, 40, "Q4_K_M", "Qwen2.5 14B Instruct, Q4_K_M 量化", 14),
|
||||
# Qwen2.5-32B-Instruct
|
||||
("Qwen2.5-32B-Instruct", "Qwen2.5-32B-Instruct (Q4_K_M)", 19.5, 64, 5120, 8, 128, 64, "Q4_K_M", "Qwen2.5 32B Instruct, Q4_K_M 量化", 15),
|
||||
("Qwen2.5-32B-Instruct", "Qwen2.5-32B-Instruct (Q8_0)", 32.0, 64, 5120, 8, 128, 64, "Q8_0", "Qwen2.5 32B Instruct, Q8_0 量化", 16),
|
||||
# Qwen2.5-72B-Instruct
|
||||
("Qwen2.5-72B-Instruct", "Qwen2.5-72B-Instruct (Q4_K_M)", 42.0, 80, 8192, 8, 128, 64, "Q4_K_M", "Qwen2.5 72B Instruct, Q4_K_M 量化", 17),
|
||||
("Qwen2.5-72B-Instruct", "Qwen2.5-72B-Instruct (Q8_0)", 75.0, 80, 8192, 8, 128, 64, "Q8_0", "Qwen2.5 72B Instruct, Q8_0 量化", 18),
|
||||
# DeepSeek-V2-Chat
|
||||
("DeepSeek-V2-Chat", "DeepSeek-V2-Chat (Q4_K_M)", 23.0, 60, 5120, 8, 128, 60, "Q4_K_M", "DeepSeek V2 Chat, Q4_K_M 量化", 19),
|
||||
# DeepSeek-V2.5-Chat
|
||||
("DeepSeek-V2.5-Chat", "DeepSeek-V2.5-Chat (Q4_K_M)", 23.0, 60, 5120, 8, 128, 60, "Q4_K_M", "DeepSeek V2.5 Chat, Q4_K_M 量化", 20),
|
||||
# DeepSeek-R1-Distill-Qwen-32B
|
||||
("DeepSeek-R1-Distill-Qwen-32B", "DeepSeek-R1-Distill-Qwen-32B (Q4_K_M)", 19.5, 64, 5120, 8, 128, 64, "Q4_K_M", "DeepSeek R1 Distill Qwen 32B, Q4_K_M 量化", 21),
|
||||
("DeepSeek-R1-Distill-Qwen-32B", "DeepSeek-R1-Distill-Qwen-32B (Q8_0)", 32.0, 64, 5120, 8, 128, 64, "Q8_0", "DeepSeek R1 Distill Qwen 32B, Q8_0 量化", 22),
|
||||
# DeepSeek-R1-Distill-Llama-70B
|
||||
("DeepSeek-R1-Distill-Llama-70B", "DeepSeek-R1-Distill-Llama-70B (Q4_K_M)", 42.0, 80, 8192, 8, 128, 64, "Q4_K_M", "DeepSeek R1 Distill Llama 70B, Q4_K_M 量化", 23),
|
||||
("DeepSeek-R1-Distill-Llama-70B", "DeepSeek-R1-Distill-Llama-70B (Q8_0)", 75.0, 80, 8192, 8, 128, 64, "Q8_0", "DeepSeek R1 Distill Llama 70B, Q8_0 量化", 24),
|
||||
# Mistral-7B-Instruct-v0.3
|
||||
("Mistral-7B-Instruct-v0.3", "Mistral-7B-Instruct-v0.3 (Q4_K_M)", 4.4, 32, 4096, 8, 128, 32, "Q4_K_M", "Mistral 7B Instruct v0.3, Q4_K_M 量化", 25),
|
||||
("Mistral-7B-Instruct-v0.3", "Mistral-7B-Instruct-v0.3 (Q8_0)", 7.5, 32, 4096, 8, 128, 32, "Q8_0", "Mistral 7B Instruct v0.3, Q8_0 量化", 26),
|
||||
# Mixtral-8x7B-Instruct
|
||||
("Mixtral-8x7B-Instruct", "Mixtral-8x7B-Instruct (Q4_K_M)", 26.0, 32, 4096, 8, 128, 32, "Q4_K_M", "Mixtral 8x7B Instruct, Q4_K_M 量化", 27),
|
||||
# Gemma-2-9B-It
|
||||
("Gemma-2-9B-It", "Gemma-2-9B-It (Q4_K_M)", 5.4, 42, 3584, 4, 256, 14, "Q4_K_M", "Google Gemma 2 9B It, Q4_K_M 量化", 28),
|
||||
# Gemma-2-27B-It
|
||||
("Gemma-2-27B-It", "Gemma-2-27B-It (Q4_K_M)", 16.5, 46, 4608, 4, 128, 36, "Q4_K_M", "Google Gemma 2 27B It, Q4_K_M 量化", 29),
|
||||
# Phi-3-Mini-4K-Instruct
|
||||
("Phi-3-Mini-4K-Instruct", "Phi-3-Mini-4K-Instruct (Q4_K_M)", 2.5, 32, 3072, 32, 96, 32, "Q4_K_M", "Microsoft Phi-3 Mini 4K Instruct, Q4_K_M 量化", 30),
|
||||
# Phi-3-Medium-14B-Instruct
|
||||
("Phi-3-Medium-14B-Instruct", "Phi-3-Medium-14B-Instruct (Q4_K_M)", 8.4, 40, 5120, 10, 128, 40, "Q4_K_M", "Microsoft Phi-3 Medium 14B Instruct, Q4_K_M 量化", 31),
|
||||
# GLM-4-9B-Chat
|
||||
("GLM-4-9B-Chat", "GLM-4-9B-Chat (Q4_K_M)", 5.5, 40, 4096, 4, 128, 40, "Q4_K_M", "Zhipu GLM-4 9B Chat, Q4_K_M 量化", 32),
|
||||
("GLM-4-9B-Chat", "GLM-4-9B-Chat (Q8_0)", 9.0, 40, 4096, 4, 128, 40, "Q8_0", "Zhipu GLM-4 9B Chat, Q8_0 量化", 33),
|
||||
# Llama-3-8B-Instruct (ctx 8192)
|
||||
("Llama-3-8B-Instruct", "Llama-3-8B-Instruct (Q4_K_M)", 4.9, 32, 4096, 8, 128, 32, 8192, "Q4_K_M", "Meta Llama 3 8B Instruct, Q4_K_M 量化", 1),
|
||||
("Llama-3-8B-Instruct", "Llama-3-8B-Instruct (Q8_0)", 8.5, 32, 4096, 8, 128, 32, 8192, "Q8_0", "Meta Llama 3 8B Instruct, Q8_0 量化", 2),
|
||||
("Llama-3-8B-Instruct", "Llama-3-8B-Instruct (FP16)", 15.5, 32, 4096, 8, 128, 32, 8192, "FP16", "Meta Llama 3 8B Instruct, FP16", 3),
|
||||
# Llama-3-70B-Instruct (ctx 8192)
|
||||
("Llama-3-70B-Instruct", "Llama-3-70B-Instruct (Q4_K_M)", 38.5, 80, 8192, 8, 128, 64, 8192, "Q4_K_M", "Meta Llama 3 70B Instruct, Q4_K_M 量化", 4),
|
||||
("Llama-3-70B-Instruct", "Llama-3-70B-Instruct (Q8_0)", 74.0, 80, 8192, 8, 128, 64, 8192, "Q8_0", "Meta Llama 3 70B Instruct, Q8_0 量化", 5),
|
||||
("Llama-3-70B-Instruct", "Llama-3-70B-Instruct (FP16)", 138.0, 80, 8192, 8, 128, 64, 8192, "FP16", "Meta Llama 3 70B Instruct, FP16", 6),
|
||||
# Llama-3.1-8B-Instruct (ctx 131072)
|
||||
("Llama-3.1-8B-Instruct", "Llama-3.1-8B-Instruct (Q4_K_M)", 4.9, 32, 4096, 8, 128, 32, 131072, "Q4_K_M", "Meta Llama 3.1 8B Instruct, Q4_K_M 量化", 7),
|
||||
("Llama-3.1-8B-Instruct", "Llama-3.1-8B-Instruct (Q8_0)", 8.5, 32, 4096, 8, 128, 32, 131072, "Q8_0", "Meta Llama 3.1 8B Instruct, Q8_0 量化", 8),
|
||||
# Llama-3.1-70B-Instruct (ctx 131072)
|
||||
("Llama-3.1-70B-Instruct", "Llama-3.1-70B-Instruct (Q4_K_M)", 38.5, 80, 8192, 8, 128, 64, 131072, "Q4_K_M", "Meta Llama 3.1 70B Instruct, Q4_K_M 量化", 9),
|
||||
("Llama-3.1-70B-Instruct", "Llama-3.1-70B-Instruct (Q8_0)", 74.0, 80, 8192, 8, 128, 64, 131072, "Q8_0", "Meta Llama 3.1 70B Instruct, Q8_0 量化", 10),
|
||||
# Llama-3.1-405B-Instruct (ctx 131072)
|
||||
("Llama-3.1-405B-Instruct", "Llama-3.1-405B-Instruct (Q4_K_M)", 226.0, 126, 16384, 8, 128, 128, 131072, "Q4_K_M", "Meta Llama 3.1 405B Instruct, Q4_K_M 量化", 11),
|
||||
# Qwen2.5-7B-Instruct (ctx 32768)
|
||||
("Qwen2.5-7B-Instruct", "Qwen2.5-7B-Instruct (Q4_K_M)", 4.7, 28, 3584, 4, 128, 28, 32768, "Q4_K_M", "Qwen2.5 7B Instruct, Q4_K_M 量化", 12),
|
||||
("Qwen2.5-7B-Instruct", "Qwen2.5-7B-Instruct (Q8_0)", 7.6, 28, 3584, 4, 128, 28, 32768, "Q8_0", "Qwen2.5 7B Instruct, Q8_0 量化", 13),
|
||||
# Qwen2.5-14B-Instruct (ctx 32768)
|
||||
("Qwen2.5-14B-Instruct", "Qwen2.5-14B-Instruct (Q4_K_M)", 8.7, 40, 5120, 8, 128, 40, 32768, "Q4_K_M", "Qwen2.5 14B Instruct, Q4_K_M 量化", 14),
|
||||
# Qwen2.5-32B-Instruct (ctx 32768)
|
||||
("Qwen2.5-32B-Instruct", "Qwen2.5-32B-Instruct (Q4_K_M)", 19.5, 64, 5120, 8, 128, 64, 32768, "Q4_K_M", "Qwen2.5 32B Instruct, Q4_K_M 量化", 15),
|
||||
("Qwen2.5-32B-Instruct", "Qwen2.5-32B-Instruct (Q8_0)", 32.0, 64, 5120, 8, 128, 64, 32768, "Q8_0", "Qwen2.5 32B Instruct, Q8_0 量化", 16),
|
||||
# Qwen2.5-72B-Instruct (ctx 32768)
|
||||
("Qwen2.5-72B-Instruct", "Qwen2.5-72B-Instruct (Q4_K_M)", 42.0, 80, 8192, 8, 128, 64, 32768, "Q4_K_M", "Qwen2.5 72B Instruct, Q4_K_M 量化", 17),
|
||||
("Qwen2.5-72B-Instruct", "Qwen2.5-72B-Instruct (Q8_0)", 75.0, 80, 8192, 8, 128, 64, 32768, "Q8_0", "Qwen2.5 72B Instruct, Q8_0 量化", 18),
|
||||
# DeepSeek-V2-Chat (ctx 4096)
|
||||
("DeepSeek-V2-Chat", "DeepSeek-V2-Chat (Q4_K_M)", 23.0, 60, 5120, 8, 128, 60, 4096, "Q4_K_M", "DeepSeek V2 Chat, Q4_K_M 量化", 19),
|
||||
# DeepSeek-V2.5-Chat (ctx 4096)
|
||||
("DeepSeek-V2.5-Chat", "DeepSeek-V2.5-Chat (Q4_K_M)", 23.0, 60, 5120, 8, 128, 60, 4096, "Q4_K_M", "DeepSeek V2.5 Chat, Q4_K_M 量化", 20),
|
||||
# DeepSeek-R1-Distill-Qwen-32B (ctx 131072)
|
||||
("DeepSeek-R1-Distill-Qwen-32B", "DeepSeek-R1-Distill-Qwen-32B (Q4_K_M)", 19.5, 64, 5120, 8, 128, 64, 131072, "Q4_K_M", "DeepSeek R1 Distill Qwen 32B, Q4_K_M 量化", 21),
|
||||
("DeepSeek-R1-Distill-Qwen-32B", "DeepSeek-R1-Distill-Qwen-32B (Q8_0)", 32.0, 64, 5120, 8, 128, 64, 131072, "Q8_0", "DeepSeek R1 Distill Qwen 32B, Q8_0 量化", 22),
|
||||
# DeepSeek-R1-Distill-Llama-70B (ctx 131072)
|
||||
("DeepSeek-R1-Distill-Llama-70B", "DeepSeek-R1-Distill-Llama-70B (Q4_K_M)", 42.0, 80, 8192, 8, 128, 64, 131072, "Q4_K_M", "DeepSeek R1 Distill Llama 70B, Q4_K_M 量化", 23),
|
||||
("DeepSeek-R1-Distill-Llama-70B", "DeepSeek-R1-Distill-Llama-70B (Q8_0)", 75.0, 80, 8192, 8, 128, 64, 131072, "Q8_0", "DeepSeek R1 Distill Llama 70B, Q8_0 量化", 24),
|
||||
# Mistral-7B-Instruct-v0.3 (ctx 32768)
|
||||
("Mistral-7B-Instruct-v0.3", "Mistral-7B-Instruct-v0.3 (Q4_K_M)", 4.4, 32, 4096, 8, 128, 32, 32768, "Q4_K_M", "Mistral 7B Instruct v0.3, Q4_K_M 量化", 25),
|
||||
("Mistral-7B-Instruct-v0.3", "Mistral-7B-Instruct-v0.3 (Q8_0)", 7.5, 32, 4096, 8, 128, 32, 32768, "Q8_0", "Mistral 7B Instruct v0.3, Q8_0 量化", 26),
|
||||
# Mixtral-8x7B-Instruct (ctx 32768)
|
||||
("Mixtral-8x7B-Instruct", "Mixtral-8x7B-Instruct (Q4_K_M)", 26.0, 32, 4096, 8, 128, 32, 32768, "Q4_K_M", "Mixtral 8x7B Instruct, Q4_K_M 量化", 27),
|
||||
# Gemma-2-9B-It (ctx 8192)
|
||||
("Gemma-2-9B-It", "Gemma-2-9B-It (Q4_K_M)", 5.4, 42, 3584, 4, 256, 14, 8192, "Q4_K_M", "Google Gemma 2 9B It, Q4_K_M 量化", 28),
|
||||
# Gemma-2-27B-It (ctx 8192)
|
||||
("Gemma-2-27B-It", "Gemma-2-27B-It (Q4_K_M)", 16.5, 46, 4608, 4, 128, 36, 8192, "Q4_K_M", "Google Gemma 2 27B It, Q4_K_M 量化", 29),
|
||||
# Phi-3-Mini-4K-Instruct (ctx 4096)
|
||||
("Phi-3-Mini-4K-Instruct", "Phi-3-Mini-4K-Instruct (Q4_K_M)", 2.5, 32, 3072, 32, 96, 32, 4096, "Q4_K_M", "Microsoft Phi-3 Mini 4K Instruct, Q4_K_M 量化", 30),
|
||||
# Phi-3-Medium-14B-Instruct (ctx 14336)
|
||||
("Phi-3-Medium-14B-Instruct", "Phi-3-Medium-14B-Instruct (Q4_K_M)", 8.4, 40, 5120, 10, 128, 40, 14336, "Q4_K_M", "Microsoft Phi-3 Medium 14B Instruct, Q4_K_M 量化", 31),
|
||||
# GLM-4-9B-Chat (ctx 131072)
|
||||
("GLM-4-9B-Chat", "GLM-4-9B-Chat (Q4_K_M)", 5.5, 40, 4096, 4, 128, 40, 131072, "Q4_K_M", "Zhipu GLM-4 9B Chat, Q4_K_M 量化", 32),
|
||||
("GLM-4-9B-Chat", "GLM-4-9B-Chat (Q8_0)", 9.0, 40, 4096, 4, 128, 40, 131072, "Q8_0", "Zhipu GLM-4 9B Chat, Q8_0 量化", 33),
|
||||
]
|
||||
|
||||
for m in default_models:
|
||||
c.execute('''INSERT INTO models (base_model, name, size_gb, layers, embd, kv_heads, head_dim, attention_heads, quant, description, sort_order)
|
||||
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)''', m)
|
||||
c.execute('''INSERT INTO models (base_model, name, size_gb, layers, embd, kv_heads, head_dim, attention_heads, default_ctx, quant, description, sort_order)
|
||||
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)''', m)
|
||||
|
||||
# ===== Default settings =====
|
||||
c.execute("INSERT OR IGNORE INTO settings (key, value) VALUES (?, ?)", ("admin_password", "admin123"))
|
||||
@@ -385,7 +386,7 @@ def insert_default_data(conn):
|
||||
c.execute("INSERT OR IGNORE INTO settings (key, value) VALUES (?, ?)", ("llm_api_url", ""))
|
||||
c.execute("INSERT OR IGNORE INTO settings (key, value) VALUES (?, ?)", ("llm_api_key", ""))
|
||||
c.execute("INSERT OR IGNORE INTO settings (key, value) VALUES (?, ?)", ("llm_api_model", ""))
|
||||
c.execute("INSERT OR IGNORE INTO settings (key, value) VALUES (?, ?)", ("llm_system_prompt", "你是一个llama.cpp命令行参数解析助手。用户会用自然语言描述他想运行的模型和参数配置，你需要将其解析为JSON格式的参数。\n\n可用的参数键包括: model, ctx_size, n_gpu_layers, threads, batch_size, temperature, top_k, top_p, flash_attn, port, host, parallel, split_mode, mlock, numa, repeat_penalty, presence_penalty, frequency_penalty, seed, min_p, typical, mirostat, mirostat_lr, mirostat_ent。\n\nGPU型号会通过 _gpu_name 字段返回，GPU数量通过 _gpu_count 返回。\n\n只返回JSON，不要其他文本。"))
|
||||
c.execute("INSERT OR IGNORE INTO settings (key, value) VALUES (?, ?)", ("nl_default_text", "用自然语言描述你想要的配置，支持多行输入。\n例如：\n用RTX 4090跑Llama-3-70B，上下文8192\n温度0.7，开启flash attention\n端口设为8080"))
|
||||
|
||||
conn.commit()
|
||||
|
||||
|
||||
Reference in New Issue
Block a user