FEATURE: add a SambaNova LLM provider (#797)

Note, at the moment the context window is quite small, it is
mainly useful as a helper backend or hyde generator

app/models/ai_api_audit_log.rb

@@ -12,6 +12,7 @@ class AiApiAuditLog < ActiveRecord::Base
     Vllm = 5
     Cohere = 6
     Ollama = 7
+    SambaNova = 8
   end
 end
 

config/locales/client.en.yml

@@ -243,7 +243,7 @@ en:
         confirm_delete: Are you sure you want to delete this model?
         delete: Delete
         seeded_warning: "This model is pre-configured on your site and cannot be edited."
-        in_use_warning: 
+        in_use_warning:
           one: "This model is currently used by the %{settings} setting. If misconfigured, the feature won't work as expected."
           other: "This model is currently used by the following settings: %{settings}. If misconfigured, features won't work as expected. "
 
@@ -275,6 +275,7 @@ en:
           azure: "Azure"
           ollama: "Ollama"
           CDCK: "CDCK"
+          samba_nova: "SambaNova"
 
         provider_fields:
           access_key_id: "AWS Bedrock Access key ID"

lib/completions/dialects/open_ai_compatible.rb

@@ -85,15 +85,14 @@ module DiscourseAi
           encoded_uploads = prompt.encoded_uploads(message)
           return content if encoded_uploads.blank?
 
-          content_w_imgs =
+          encoded_uploads.reduce([{ type: "text", text: message[:content] }]) do |memo, details|
-            encoded_uploads.reduce([{ type: "text", text: message[:content] }]) do |memo, details|
+            memo << {
-              memo << {
+              type: "image_url",
-                type: "image_url",
+              image_url: {
-                image_url: {
+                url: "data:#{details[:mime_type]};base64,#{details[:base64]}",
-                  url: "data:#{details[:mime_type]};base64,#{details[:base64]}",
+              },
-                },
+            }
-              }
+          end
-            end
         end
       end
     end

lib/completions/endpoints/base.rb

@@ -17,6 +17,7 @@ module DiscourseAi
               DiscourseAi::Completions::Endpoints::Vllm,
               DiscourseAi::Completions::Endpoints::Anthropic,
               DiscourseAi::Completions::Endpoints::Cohere,
+              DiscourseAi::Completions::Endpoints::SambaNova,
             ]
 
             endpoints << DiscourseAi::Completions::Endpoints::Ollama if Rails.env.development?

lib/completions/endpoints/samba_nova.rb

@@ -0,0 +1,84 @@
+# frozen_string_literal: true
+
+module DiscourseAi
+  module Completions
+    module Endpoints
+      class SambaNova < Base
+        def self.can_contact?(model_provider)
+          model_provider == "samba_nova"
+        end
+
+        def normalize_model_params(model_params)
+          model_params = model_params.dup
+
+          # max_tokens, temperature are already supported
+          if model_params[:stop_sequences]
+            model_params[:stop] = model_params.delete(:stop_sequences)
+          end
+
+          model_params
+        end
+
+        def default_options
+          { model: llm_model.name }
+        end
+
+        def provider_id
+          AiApiAuditLog::Provider::SambaNova
+        end
+
+        private
+
+        def model_uri
+          URI(llm_model.url)
+        end
+
+        def prepare_payload(prompt, model_params, dialect)
+          payload = default_options.merge(model_params).merge(messages: prompt)
+
+          payload[:stream] = true if @streaming_mode
+
+          payload
+        end
+
+        def prepare_request(payload)
+          headers = { "Content-Type" => "application/json" }
+          api_key = llm_model.api_key
+
+          headers["Authorization"] = "Bearer #{api_key}"
+
+          Net::HTTP::Post.new(model_uri, headers).tap { |r| r.body = payload }
+        end
+
+        def final_log_update(log)
+          log.request_tokens = @prompt_tokens if @prompt_tokens
+          log.response_tokens = @completion_tokens if @completion_tokens
+        end
+
+        def extract_completion_from(response_raw)
+          json = JSON.parse(response_raw, symbolize_names: true)
+
+          if @streaming_mode
+            @prompt_tokens ||= json.dig(:usage, :prompt_tokens)
+            @completion_tokens ||= json.dig(:usage, :completion_tokens)
+          end
+
+          parsed = json.dig(:choices, 0)
+          return if !parsed
+
+          @streaming_mode ? parsed.dig(:delta, :content) : parsed.dig(:message, :content)
+        end
+
+        def partials_from(decoded_chunk)
+          decoded_chunk
+            .split("\n")
+            .map do |line|
+              data = line.split("data: ", 2)[1]
+              data == "[DONE]" ? nil : data
+            end
+            .compact
+        end
+      end
+    end
+  end
+end

lib/completions/llm.rb

@@ -76,7 +76,17 @@ module DiscourseAi
         end
 
         def provider_names
-          providers = %w[aws_bedrock anthropic vllm hugging_face cohere open_ai google azure]
+          providers = %w[
+            aws_bedrock
+            anthropic
+            vllm
+            hugging_face
+            cohere
+            open_ai
+            google
+            azure
+            samba_nova
+          ]
           if !Rails.env.production?
             providers << "fake"
             providers << "ollama"

spec/fabricators/llm_model_fabricator.rb

@@ -71,3 +71,11 @@ Fabricator(:cohere_model, from: :llm_model) do
   api_key "ABC"
   url "https://api.cohere.ai/v1/chat"
 end
+
+Fabricator(:samba_nova_model, from: :llm_model) do
+  display_name "Samba Nova"
+  name "samba-nova"
+  provider "samba_nova"
+  api_key "ABC"
+  url "https://api.sambanova.ai/v1/chat/completions"
+end

spec/lib/completions/endpoints/samba_nova_spec.rb

@@ -0,0 +1,47 @@
+# frozen_string_literal: true
+
+RSpec.describe DiscourseAi::Completions::Endpoints::SambaNova do
+  fab!(:llm_model) { Fabricate(:samba_nova_model) }
+  let(:llm) { llm_model.to_llm }
+
+  it "can stream completions" do
+    body = <<~PARTS
+    data: {"id": "4c5e4a44-e847-467d-b9cd-d2f6530678cd", "object": "chat.completion.chunk", "created": 1721336361, "model": "llama3-8b", "system_fingerprint": "fastcoe", "choices": [{"index": 0, "delta": {"content": "I am a bot"}, "logprobs": null, "finish_reason": null}]}
+
+data: {"id": "4c5e4a44-e847-467d-b9cd-d2f6530678cd", "object": "chat.completion.chunk", "created": 1721336361, "model": "llama3-8b", "system_fingerprint": "fastcoe", "choices": [], "usage": {"is_last_response": true, "total_tokens": 62, "prompt_tokens": 21, "completion_tokens": 41, "time_to_first_token": 0.09152531623840332, "end_time": 1721336361.582011, "start_time": 1721336361.413994, "total_latency": 0.16801691055297852, "total_tokens_per_sec": 369.010475171488, "completion_tokens_per_sec": 244.02305616179046, "completion_tokens_after_first_per_sec": 522.9332759819093, "completion_tokens_after_first_per_sec_first_ten": 1016.0004844667837}}
+
+data: [DONE]
+    PARTS
+
+    stub_request(:post, "https://api.sambanova.ai/v1/chat/completions").with(
+      body:
+        "{\"model\":\"samba-nova\",\"messages\":[{\"role\":\"system\",\"content\":\"You are a helpful bot\"},{\"role\":\"user\",\"content\":\"who are you?\"}],\"stream\":true}",
+      headers: {
+        "Authorization" => "Bearer ABC",
+        "Content-Type" => "application/json",
+      },
+    ).to_return(status: 200, body: body, headers: {})
+
+    response = +""
+    llm.generate("who are you?", user: Discourse.system_user) { |partial| response << partial }
+
+    expect(response).to eq("I am a bot")
+  end
+
+  it "can perform regular completions" do
+    body = { choices: [message: { content: "I am a bot" }] }.to_json
+
+    stub_request(:post, "https://api.sambanova.ai/v1/chat/completions").with(
+      body:
+        "{\"model\":\"samba-nova\",\"messages\":[{\"role\":\"system\",\"content\":\"You are a helpful bot\"},{\"role\":\"user\",\"content\":\"who are you?\"}]}",
+      headers: {
+        "Authorization" => "Bearer ABC",
+        "Content-Type" => "application/json",
+      },
+    ).to_return(status: 200, body: body, headers: {})
+
+    response = llm.generate("who are you?", user: Discourse.system_user)
+
+    expect(response).to eq("I am a bot")
+  end
+end