Add DeepSeek models (#147)

CHANGELOG.md

@@ -7,6 +7,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Added
+- Added support for [DeepSeek models](https://platform.deepseek.com/docs) via the `dschat` and `dscode` aliases. You can set the `DEEPSEEK_API_KEY` environment variable to your DeepSeek API key.
 
 ### Fixed
 

Project.toml

@@ -1,7 +1,7 @@
 name = "PromptingTools"
 uuid = "670122d1-24a8-4d70-bfce-740807c42192"
 authors = ["J S @svilupp and contributors"]
-version = "0.23.0"
+version = "0.24.0"
 
 [deps]
 AbstractTrees = "1520ce14-60c1-5f80-bbc7-55ef81b5835c"

docs/src/frequently_asked_questions.md

@@ -415,3 +415,58 @@ Fine-tuning is a powerful technique to adapt a model to your specific use case (
 2. Once the finetuning time comes, create a bundle of ShareGPT-formatted conversations (common finetuning format) in a single `.jsonl` file. Use `PT.save_conversations("dataset.jsonl", [conversation1, conversation2, ...])` (notice that plural "conversationS" in the function name).
 
 For an example of an end-to-end finetuning process, check out our sister project [JuliaLLMLeaderboard Finetuning experiment](https://github.com/svilupp/Julia-LLM-Leaderboard/blob/main/experiments/cheater-7b-finetune/README.md). It shows the process of finetuning for half a dollar with [JarvisLabs.ai](https://jarvislabs.ai/templates/axolotl) and [Axolotl](https://github.com/OpenAccess-AI-Collective/axolotl).
+
+## Can I see how my prompt is rendered / what is sent to the API?
+
+Yes, there are two ways. 
+1) "dry run", where the `ai*` function will return the prompt rendered in the style of the selected API provider
+2) "partial render", for provider-agnostic purposes, you can run only the first step of the rendering pipeline to see the messages that will be sent (but formatted as `SystemMessage` and `UserMessage`), which is easy to read and work with
+
+1) Dry Run
+
+Add kwargs `dry_run` and `return_all` to see what could have been sent to the API to your `ai*` functions (without `return_all` there is nothing to show you).
+
+Example for OpenAI:
+```julia
+dry_conv = aigenerate(:BlankSystemUser; system = "I exist", user = "say hi",
+    model = "lngpt3t", return_all = true, dry_run = true)
+```
+
+```plaintext
+2-element Vector{Dict{String, Any}}:
+ Dict("role" => "system", "content" => "I exist")
+ Dict("role" => "user", "content" => "say hi")
+```
+
+2) Partial Render
+
+Personally, I prefer to see the pretty formatting of PromptingTools *Messages. 
+To see what will be sent to the model, you can `render` only the first stage of the rendering pipeline with schema `NoSchema()` (it merely does the variable replacements and creates the necessary messages). It's shared by all the schema/providers.
+
+```julia
+PT.render(PT.NoSchema(), "say hi, {{name}}"; name="John")
+```
+
+```plaintext
+2-element Vector{PromptingTools.AbstractMessage}:
+ PromptingTools.SystemMessage("Act as a helpful AI assistant")
+ PromptingTools.UserMessage("say hi, John")
+```
+
+What about the prompt templates?
+Prompt templates have an extra pre-rendering step that expands the symbolic `:name` (understood by PromptingTools as a reference to `AITemplate(:name)`) into a vector of Messages.
+
+```julia
+# expand the template into messages
+tpl = PT.render(AITemplate(:BlankSystemUser))
+PT.render(PT.NoSchema(), tpl; system = "I exist", user = "say hi")
+# replace any variables, etc.
+```
+
+```plaintext
+2-element Vector{PromptingTools.AbstractMessage}:
+ PromptingTools.SystemMessage("I exist")
+ PromptingTools.UserMessage("say hi")
+```
+
+For more information about the rendering pipeline and examples refer to [Walkthrough Example for aigenerate](@ref).

src/llm_interface.jl

@@ -192,6 +192,20 @@ Requires one environment variables to be set:
 """
 struct GroqOpenAISchema <: AbstractOpenAISchema end
 
+"""
+    DeepSeekOpenAISchema
+
+Schema to call the [DeepSeek](https://platform.deepseek.com/docs) API.
+
+Links:
+- [Get your API key](https://platform.deepseek.com/api_keys)
+- [API Reference](https://platform.deepseek.com/docs)
+
+Requires one environment variables to be set:
+- `DEEPSEEK_API_KEY`: Your API key (often starts with "sk-...")
+"""
+struct DeepSeekOpenAISchema <: AbstractOpenAISchema end
+
 abstract type AbstractOllamaSchema <: AbstractPromptSchema end
 
 """

src/llm_openai.jl

@@ -205,6 +205,19 @@ function OpenAI.create_chat(schema::GroqOpenAISchema,
         base_url = url)
     OpenAI.create_chat(provider, model, conversation; kwargs...)
 end
+function OpenAI.create_chat(schema::DeepSeekOpenAISchema,
+        api_key::AbstractString,
+        model::AbstractString,
+        conversation;
+        url::String = "https://api.deepseek.com/v1",
+        kwargs...)
+    # Build the corresponding provider object
+    # try to override provided api_key because the default is OpenAI key
+    provider = CustomProvider(;
+        api_key = isempty(DEEPSEEK_API_KEY) ? api_key : DEEPSEEK_API_KEY,
+        base_url = url)
+    OpenAI.create_chat(provider, model, conversation; kwargs...)
+end
 function OpenAI.create_chat(schema::DatabricksOpenAISchema,
         api_key::AbstractString,
         model::AbstractString,

src/user_preferences.jl

@@ -20,6 +20,7 @@ Check your preferences by calling `get_preferences(key::String)`.
 - `ANTHROPIC_API_KEY`: The API key for the Anthropic API. Get yours from [here](https://www.anthropic.com/).
 - `VOYAGE_API_KEY`: The API key for the Voyage API. Free tier is upto 50M tokens! Get yours from [here](https://dash.voyageai.com/api-keys).
 - `GROQ_API_KEY`: The API key for the Groq API. Free in beta! Get yours from [here](https://console.groq.com/keys).
+- `DEEPSEEK_API_KEY`: The API key for the DeepSeek API. Get \$5 credit when you join. Get yours from [here](https://platform.deepseek.com/api_keys).
 - `MODEL_CHAT`: The default model to use for aigenerate and most ai* calls. See `MODEL_REGISTRY` for a list of available models or define your own.
 - `MODEL_EMBEDDING`: The default model to use for aiembed (embedding documents). See `MODEL_REGISTRY` for a list of available models or define your own.
 - `PROMPT_SCHEMA`: The default prompt schema to use for aigenerate and most ai* calls (if not specified in `MODEL_REGISTRY`). Set as a string, eg, `"OpenAISchema"`.
@@ -46,6 +47,7 @@ Define your `register_model!()` calls in your `startup.jl` file to make them ava
 - `ANTHROPIC_API_KEY`: The API key for the Anthropic API. Get yours from [here](https://www.anthropic.com/).
 - `VOYAGE_API_KEY`: The API key for the Voyage API. Free tier is upto 50M tokens! Get yours from [here](https://dash.voyageai.com/api-keys).
 - `GROQ_API_KEY`: The API key for the Groq API. Free in beta! Get yours from [here](https://console.groq.com/keys).
+- `DEEPSEEK_API_KEY`: The API key for the DeepSeek API. Get \$5 credit when you join. Get yours from [here](https://platform.deepseek.com/api_keys).
 
 Preferences.jl takes priority over ENV variables, so if you set a preference, it will take precedence over the ENV variable.
 
@@ -64,6 +66,7 @@ const ALLOWED_PREFERENCES = ["MISTRALAI_API_KEY",
     "ANTHROPIC_API_KEY",
     "VOYAGE_API_KEY",
     "GROQ_API_KEY",
+    "DEEPSEEK_API_KEY",
     "MODEL_CHAT",
     "MODEL_EMBEDDING",
     "MODEL_ALIASES",
@@ -179,6 +182,10 @@ _temp = get(ENV, "GROQ_API_KEY", "")
 const GROQ_API_KEY::String = @load_preference("GROQ_API_KEY",
     default=_temp);
 
+_temp = get(ENV, "DEEPSEEK_API_KEY", "")
+const DEEPSEEK_API_KEY::String = @load_preference("DEEPSEEK_API_KEY",
+    default=_temp);
+
 _temp = get(ENV, "LOCAL_SERVER", "http://localhost:10897/v1")
 ## Address of the local server
 const LOCAL_SERVER::String = @load_preference("LOCAL_SERVER",
@@ -341,7 +348,10 @@ aliases = merge(
         "gl3" => "llama3-8b-8192",
         "gllama370" => "llama3-70b-8192",
         "gl70" => "llama3-70b-8192",
-        "gmixtral" => "mixtral-8x7b-32768"
+        "gmixtral" => "mixtral-8x7b-32768",
+        ## DeepSeek
+        "dschat" => "deepseek-chat",
+        "dscode" => "deepseek-coder"
     ),
     ## Load aliases from preferences as well
     @load_preference("MODEL_ALIASES", default=Dict{String, String}()))
@@ -646,7 +656,17 @@ registry = Dict{String, ModelSpec}(
         GroqOpenAISchema(),
         2.7e-7,
         2.7e-7,
-        "Mistral.ai Mixtral 8x7b, hosted by Groq. Max 32K context. See details [here](https://console.groq.com/docs/models)")
+        "Mistral.ai Mixtral 8x7b, hosted by Groq. Max 32K context. See details [here](https://console.groq.com/docs/models)"),
+    "deepseek-chat" => ModelSpec("deepseek-chat",
+        DeepSeekOpenAISchema(),
+        1.4e-7,
+        2.8e-7,
+        "Deepseek.com-hosted DeepSeekV2 model. Max 32K context. See details [here](https://platform.deepseek.com/docs)"),
+    "deepseek-coder" => ModelSpec("deepseek-coder",
+        DeepSeekOpenAISchema(),
+        1.4e-7,
+        2.8e-7,
+        "Deepseek.com-hosted coding model. Max 16K context. See details [here](https://platform.deepseek.com/docs)")
 )
 
 """