feat: add air for development and start on support for claude v3

.air.toml

@@ -0,0 +1,46 @@
+root = "."
+testdata_dir = "testdata"
+tmp_dir = "tmp"
+
+[build]
+  args_bin = ["server"]
+  bin = "./tmp/main"
+  cmd = "go build -o ./tmp/main ."
+  delay = 1000
+  exclude_dir = ["assets", "tmp", "vendor", "testdata"]
+  exclude_file = []
+  exclude_regex = ["_test.go"]
+  exclude_unchanged = false
+  follow_symlink = false
+  full_bin = ""
+  include_dir = []
+  include_ext = ["go", "tpl", "tmpl", "html"]
+  include_file = []
+  kill_delay = "0s"
+  log = "build-errors.log"
+  poll = false
+  poll_interval = 0
+  post_cmd = []
+  pre_cmd = []
+  rerun = false
+  rerun_delay = 500
+  send_interrupt = false
+  stop_on_error = false
+
+[color]
+  app = ""
+  build = "yellow"
+  main = "magenta"
+  runner = "green"
+  watcher = "cyan"
+
+[log]
+  main_only = false
+  time = false
+
+[misc]
+  clean_on_exit = false
+
+[screen]
+  clear_on_rebuild = false
+  keep_scroll = true

.gitignore

@@ -0,0 +1,2 @@
+tmp/*
+jenn-ai

internal/bedrock/bedrock.go

@@ -84,6 +84,26 @@ func (m *AWSModelConfig) constructPayload(message string) ([]byte, error) {
 			TopP:              m.TopP,
 			TopK:              m.TopK,
 		}
+		// TODO: work on v3
+		//body := models.ClaudeMessagesInput{
+		//	AnthropicVersion: "bedrock-2023-05-31",
+		//	Messages: []models.ClaudeMessage{
+		//		{
+		//			Role: "user",
+		//			Content: []models.ClaudeContent{
+		//				{
+		//					Type: "text",
+		//					Text: message,
+		//				},
+		//			},
+		//		},
+		//	},
+		//	MaxTokens:   m.MaxTokens,
+		//	Temperature: m.Temperature,
+		//	TopP:        m.TopP,
+		//	TopK:        m.TopK,
+		//}
+
 		payload, err := json.Marshal(body)
 		if err != nil {
 			return []byte{}, err

internal/bedrock/models/anthropic.go

@@ -1,18 +1,60 @@
 // Package models contains structs on model requests/responses
 package models
 
-// ClaudeModelInputs is needed for the request
+// ClaudeModelInputs is needed to marshal the legacy request type
+// Supported Models: claude-instant-v1, claude-v1, claude-v2, claude-v2:1
 type ClaudeModelInputs struct {
-	Prompt            string  `json:"prompt"`
-	MaxTokensToSample int     `json:"max_tokens_to_sample"`
-	Temperature       float64 `json:"temperature,omitempty"`
-	TopP              float64 `json:"top_p,omitempty"`
-	TopK              int     `json:"top_k,omitempty"`
+	Prompt            string  `json:"prompt"`                // The prompt that you want Claude to complete. For proper response generation you need to format your prompt using alternating \n\nHuman: and \n\nAssistant: conversational turns.
+	MaxTokensToSample int     `json:"max_tokens_to_sample"`  // The maximum number of tokens to generate before stopping.
+	Temperature       float64 `json:"temperature,omitempty"` // The amount of randomness injected into the response.
+	TopP              float64 `json:"top_p,omitempty"`       // Use nucleus sampling.
+	TopK              int     `json:"top_k,omitempty"`       // Only sample from the top K options for each subsequent token.
 }
 
-// ClaudeModelOutputs contains the response
+// ClaudeModelOutputs is needed to unmarshal the legacy response
+// Supported Models: claude-instant-v1, claude-v1, claude-v2, claude-v2:1
 type ClaudeModelOutputs struct {
-	Completion string `json:"completion,omitempty"`
-	StopReason string `json:"stop_reason,omitempty"`
-	Stop       string `json:"stop,omitempty"`
+	Completion string `json:"completion,omitempty"`  // The resulting completion up to and excluding the stop sequences.
+	StopReason string `json:"stop_reason,omitempty"` // The reason why the model stopped generating the response.
+	Stop       string `json:"stop,omitempty"`        // contains the stop sequence that signalled the model to stop generating text.
+}
+
+// ClaudeMessagesInput is needed to marshal the new request type
+// Supported Models: claude-instant-v1.2, claude-v2, claude-v2.1, claude-v3
+type ClaudeMessagesInput struct {
+	AnthropicVersion string          `json:"anthropic_version"`        // The anthropic version. The value must be bedrock-2023-05-31.
+	Messages         []ClaudeMessage `json:"messages"`                 // The input messages.
+	MaxTokens        int             `json:"max_tokens"`               // The maximum number of tokens to generate before stopping.
+	Temperature      float64         `json:"temperature"`              // The amount of randomness injected into the response.
+	TopK             int             `json:"top_k"`                    // Only sample from the top K options for each subsequent token.
+	TopP             float64         `json:"top_p"`                    // Use nucleus sampling.
+	System           string          `json:"system,omitempty"`         // The system prompt for the request. Which provides context, instructions, and guidelines to Claude before presenting it with a question or task.
+	StopSequences    []string        `json:"stop_sequences,omitempty"` // The stop sequences that signal the model to stop generating text.
+}
+
+// ClaudeMessage contains messages
+type ClaudeMessage struct {
+	Role    string          `json:"role"`    // The role of the conversation turn. Valid values are user and assistant.
+	Content []ClaudeContent `json:"content"` // The content of the conversation turn.
+}
+
+// ClaudeContent contains content
+type ClaudeContent struct {
+	Type   string         `json:"type"`             // The type of the content. Valid values are image and text.
+	Text   string         `json:"text,omitempty"`   // The text content.
+	Source []ClaudeSource `json:"source,omitempty"` // The content of the conversation turn.
+}
+
+// ClaudeSource contains source
+type ClaudeSource struct {
+	Type      string `json:"type"`       // The encoding type for the image. You can specify base64.
+	MediaType string `json:"media_type"` // The type of the image. You can specify the following image formats.
+	Data      string `json:"data"`       // The base64 encoded image bytes for the image. The maximum image size is 3.75MB.
+}
+
+// ClaudeMessagesOutput is needed to unmarshal the new request type
+// Supported Models: claude-instant-v1.2, claude-v2, claude-v2.1, claude-v3
+type ClaudeMessagesOutput struct {
+	Content    []ClaudeContent `json:"content"`     // The content generated by the model.
+	StopReason string          `json:"stop_reason"` // The reason why Anthropic Claude stopped generating the response.
 }

internal/bedrock/models/cohere.go

@@ -3,24 +3,24 @@ package models
 
 // CommandModelInput contains the request
 type CommandModelInput struct {
-	Prompt            string   `json:"prompt"`
-	Temperature       float64  `json:"temperature"`
-	TopP              float64  `json:"p"`
-	TopK              int      `json:"k"`
-	MaxTokensToSample int      `json:"max_tokens"`
-	StopSequences     []string `json:"stop_sequences"`
-	ReturnLiklihoods  string   `json:"return_likelihoods"`
-	Stream            bool     `json:"stream"`
-	NumGenerations    int      `json:"num_generations"`
+	Prompt            string   `json:"prompt"`             // The input text that serves as the starting point for generating the response.
+	Temperature       float64  `json:"temperature"`        // Use a lower value to decrease randomness in the response.
+	TopP              float64  `json:"p"`                  // Use a lower value to ignore less probable options. Set to 0 or 1.0 to disable.
+	TopK              int      `json:"k"`                  // Specify the number of token choices the model uses to generate the next token.
+	MaxTokensToSample int      `json:"max_tokens"`         // Specify the maximum number of tokens to use in the generated response.
+	StopSequences     []string `json:"stop_sequences"`     // Configure up to four sequences that the model recognizes.
+	ReturnLiklihoods  string   `json:"return_likelihoods"` // Specify how and if the token likelihoods are returned with the response.
+	Stream            bool     `json:"stream"`             // Specify true to return the response piece-by-piece in real-time and false to return the complete response after the process finishes.
+	NumGenerations    int      `json:"num_generations"`    // The maximum number of generations that the model should return.
 }
 
 // CommandModelOutput contains the response
 type CommandModelOutput struct {
-	Generations []CommandGeneration `json:"generations"`
+	Generations []CommandGeneration `json:"generations"` // A list of generated results along with the likelihoods for tokens requested.
 }
 
 // CommandGeneration contains the chunks of text
 type CommandGeneration struct {
-	ID   string `json:"id"`
-	Text string `json:"text"`
+	ID   string `json:"id"`   // An identifier for the generation.
+	Text string `json:"text"` // The generated text.
 }