feat: models(gallery): add deepseek-v2-lite

core/config/guesser.go

@@ -20,6 +20,7 @@ const (
 	ChatML
 	Mistral03
 	Gemma
+	DeepSeek2
 )
 
 type settingsConfig struct {
@@ -37,6 +38,17 @@ var defaultsSettings map[familyType]settingsConfig = map[familyType]settingsConf
 			Completion:  "{{.Input}}",
 		},
 	},
+	DeepSeek2: {
+		StopWords: []string{"<｜end▁of▁sentence｜>"},
+		TemplateConfig: TemplateConfig{
+			ChatMessage: `{{if eq .RoleName "user" -}}User: {{.Content }}
+{{ end -}}
+{{if eq .RoleName "assistant" -}}Assistant: {{.Content}}<｜end▁of▁sentence｜>{{end}}
+{{if eq .RoleName "system" -}}{{.Content}}
+{{end -}}`,
+			Chat: "{{.Input -}}\nAssistant: ",
+		},
+	},
 	LLaMa3: {
 		StopWords: []string{"<|eot_id|>"},
 		TemplateConfig: TemplateConfig{
@@ -208,8 +220,11 @@ func identifyFamily(f *gguf.GGUFFile) familyType {
 	qwen2 := arch == "qwen2"
 	phi3 := arch == "phi-3"
 	gemma := strings.HasPrefix(f.Model().Name, "gemma")
+	deepseek2 := arch == "deepseek2"
 
 	switch {
+	case deepseek2:
+		return DeepSeek2
 	case gemma:
 		return Gemma
 	case llama3:

gallery/deepseek.yaml

@@ -0,0 +1,20 @@
+---
+name: "deepseek"
+
+config_file: |
+  mmap: true
+  context_size: 8192
+  template:
+    chat_message: |-
+      {{if eq .RoleName "user" -}}User: {{.Content }}
+      {{ end -}}
+      {{if eq .RoleName "assistant" -}}Assistant: {{.Content}}<｜end▁of▁sentence｜>{{end}}
+      {{if eq .RoleName "system" -}}{{.Content}}
+      {{end -}}
+    chat: |
+      {{.Input -}}
+      Assistant: # Space is preserved for templating reasons, but line does not end with one for the linter.
+    completion: |
+      {{.Input}}
+  stopwords:
+  - '<｜end▁of▁sentence｜>'

gallery/index.yaml

@@ -1,4 +1,29 @@
 ---
+## Deepseek
+- &deepseek
+  url: "github:mudler/LocalAI/gallery/deepseek.yaml@master"
+  name: "deepseek-coder-v2-lite-instruct"
+  icon: "https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/logo.svg?raw=true"
+  license: deepseek
+  description: |
+    DeepSeek-Coder-V2, an open-source Mixture-of-Experts (MoE) code language model that achieves performance comparable to GPT4-Turbo in code-specific tasks. Specifically, DeepSeek-Coder-V2 is further pre-trained from DeepSeek-Coder-V2-Base with 6 trillion tokens sourced from a high-quality and multi-source corpus. Through this continued pre-training, DeepSeek-Coder-V2 substantially enhances the coding and mathematical reasoning capabilities of DeepSeek-Coder-V2-Base, while maintaining comparable performance in general language tasks. Compared to DeepSeek-Coder, DeepSeek-Coder-V2 demonstrates significant advancements in various aspects of code-related tasks, as well as reasoning and general capabilities. Additionally, DeepSeek-Coder-V2 expands its support for programming languages from 86 to 338, while extending the context length from 16K to 128K.
+    In standard benchmark evaluations, DeepSeek-Coder-V2 achieves superior performance compared to closed-source models such as GPT4-Turbo, Claude 3 Opus, and Gemini 1.5 Pro in coding and math benchmarks. The list of supported programming languages can be found in the paper.
+  urls:
+    - https://github.com/deepseek-ai/DeepSeek-Coder-V2/tree/main
+    - https://huggingface.co/LoneStriker/DeepSeek-Coder-V2-Lite-Instruct-GGUF
+  tags:
+    - llm
+    - gguf
+    - gpu
+    - deepseek
+    - cpu
+  overrides:
+    parameters:
+      model: DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf
+  files:
+    - filename: DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf
+      sha256: 50ec78036433265965ed1afd0667c00c71c12aa70bcf383be462cb8e159db6c0
+      uri: huggingface://LoneStriker/DeepSeek-Coder-V2-Lite-Instruct-GGUF/DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf
 ## Start QWEN2
 - &qwen2
   url: "github:mudler/LocalAI/gallery/chatml.yaml@master"