llms: extract common llms errors to shared declaration

llms/anthropic/anthropicllm.go

@@ -14,7 +14,6 @@ import (
 )
 
 var (
-	ErrEmptyResponse            = errors.New("no response")
 	ErrMissingToken             = errors.New("missing the Anthropic API key, set it in the ANTHROPIC_API_KEY environment variable")
 	ErrUnexpectedResponseLength = errors.New("unexpected length of response")
 	ErrInvalidContentType       = errors.New("invalid content type")
@@ -92,7 +91,7 @@ func (o *LLM) GenerateContent(ctx context.Context, messages []llms.MessageConten
 
 func generateCompletionsContent(ctx context.Context, o *LLM, messages []llms.MessageContent, opts *llms.CallOptions) (*llms.ContentResponse, error) {
 	if len(messages) == 0 || len(messages[0].Parts) == 0 {
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 
 	msg0 := messages[0]
@@ -153,7 +152,7 @@ func generateMessagesContent(ctx context.Context, o *LLM, messages []llms.Messag
 		return nil, fmt.Errorf("anthropic: failed to create message: %w", err)
 	}
 	if result == nil {
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 
 	choices := make([]*llms.ContentChoice, len(result.Content))

llms/anthropic/internal/anthropicclient/anthropicclient.go

@@ -16,9 +16,6 @@ const (
 	defaultModel = "claude-3-5-sonnet-20240620"
 )
 
-// ErrEmptyResponse is returned when the Anthropic API returns an empty response.
-var ErrEmptyResponse = errors.New("empty response")
-
 // Client is a client for the Anthropic API.
 type Client struct {
 	token   string

llms/bedrock/internal/bedrockclient/defaults.go

@@ -0,0 +1,7 @@
+package bedrockclient
+
+const (
+	DefaultMaxTokenLength2048 = 2048
+	DefaultMaxTokenLength512  = 512
+	DefaultMaxTokenLength20   = 20
+)

llms/bedrock/internal/bedrockclient/provider_ai21.go

@@ -81,7 +81,7 @@ func createAi21Completion(ctx context.Context, client *bedrockruntime.Client, mo
 		Prompt:        txt,
 		Temperature:   options.Temperature,
 		TopP:          options.TopP,
-		MaxTokens:     getMaxTokens(options.MaxTokens, 2048),
+		MaxTokens:     getMaxTokens(options.MaxTokens, DefaultMaxTokenLength2048),
 		StopSequences: options.StopWords,
 		CountPenalty: struct {
 			Scale float64 `json:"scale"`

llms/bedrock/internal/bedrockclient/provider_amazon.go

@@ -67,7 +67,7 @@ func createAmazonCompletion(ctx context.Context,
 	inputContent := amazonTextGenerationInput{
 		InputText: txt,
 		TextGenerationConfig: amazonTextGenerationConfigInput{
-			MaxTokens:     getMaxTokens(options.MaxTokens, 512),
+			MaxTokens:     getMaxTokens(options.MaxTokens, DefaultMaxTokenLength512),
 			TopP:          options.TopP,
 			Temperature:   options.Temperature,
 			StopSequences: options.StopWords,

llms/bedrock/internal/bedrockclient/provider_anthropic.go

@@ -134,7 +134,7 @@ func createAnthropicCompletion(ctx context.Context,
 
 	input := anthropicTextGenerationInput{
 		AnthropicVersion: AnthropicLatestVersion,
-		MaxTokens:        getMaxTokens(options.MaxTokens, 2048),
+		MaxTokens:        getMaxTokens(options.MaxTokens, DefaultMaxTokenLength2048),
 		System:           systemPrompt,
 		Messages:         inputContents,
 		Temperature:      options.Temperature,

llms/bedrock/internal/bedrockclient/provider_cohere.go

@@ -74,7 +74,7 @@ func createCohereCompletion(ctx context.Context,
 		Temperature:    options.Temperature,
 		P:              options.TopP,
 		K:              options.TopK,
-		MaxTokens:      getMaxTokens(options.MaxTokens, 20),
+		MaxTokens:      getMaxTokens(options.MaxTokens, DefaultMaxTokenLength20),
 		StopSequences:  options.StopWords,
 		NumGenerations: options.CandidateCount,
 	}

llms/bedrock/internal/bedrockclient/provider_meta.go

@@ -56,7 +56,7 @@ func createMetaCompletion(ctx context.Context,
 		Prompt:      txt,
 		Temperature: options.Temperature,
 		TopP:        options.TopP,
-		MaxGenLen:   getMaxTokens(options.MaxTokens, 512),
+		MaxGenLen:   getMaxTokens(options.MaxTokens, DefaultMaxTokenLength512),
 	}
 
 	body, err := json.Marshal(input)

llms/cloudflare/cloudflarellm.go

@@ -10,11 +10,6 @@ import (
 	"github.com/tmc/langchaingo/llms/cloudflare/internal/cloudflareclient"
 )
 
-var (
-	ErrEmptyResponse       = errors.New("no response")
-	ErrIncompleteEmbedding = errors.New("not all input got embedded")
-)
-
 // LLM is a cloudflare LLM implementation.
 type LLM struct {
 	CallbacksHandler callbacks.Handler
@@ -147,11 +142,11 @@ func (o *LLM) CreateEmbedding(ctx context.Context, inputTexts []string) ([][]flo
 	}
 
 	if len(res.Result.Data) == 0 {
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 
 	if len(inputTexts) != len(res.Result.Data) {
-		return res.Result.Data, ErrIncompleteEmbedding
+		return res.Result.Data, llms.ErrIncompleteEmbedding
 	}
 
 	return res.Result.Data, nil

llms/cloudflare/internal/cloudflareclient/api.go

@@ -37,7 +37,7 @@ func (c *Client) CreateEmbedding(ctx context.Context, texts *CreateEmbeddingRequ
 		return nil, err
 	}
 
-	if resp.StatusCode > 299 {
+	if resp.StatusCode >= http.StatusMultipleChoices {
 		return nil, fmt.Errorf("error: %s", body)
 	}
 
@@ -81,7 +81,7 @@ func (c *Client) GenerateContent(ctx context.Context, request *GenerateContentRe
 			return nil, err
 		}
 
-		if response.StatusCode > 299 {
+		if response.StatusCode >= http.StatusMultipleChoices {
 			return nil, fmt.Errorf("error: %s", body)
 		}
 
@@ -165,7 +165,7 @@ func (c *Client) Summarize(ctx context.Context, inputText string, maxLength int)
 		return nil, err
 	}
 
-	if resp.StatusCode > 299 {
+	if resp.StatusCode >= http.StatusMultipleChoices {
 		return nil, fmt.Errorf("error: %s", body)
 	}
 

llms/cohere/coherellm.go

@@ -11,8 +11,7 @@ import (
 )
 
 var (
-	ErrEmptyResponse = errors.New("no response")
-	ErrMissingToken  = errors.New("missing the COHERE_API_KEY key, set it in the COHERE_API_KEY environment variable")
+	ErrMissingToken = errors.New("missing the COHERE_API_KEY key, set it in the COHERE_API_KEY environment variable")
 
 	ErrUnexpectedResponseLength = errors.New("unexpected length of response")
 )

llms/cohere/internal/cohereclient/cohereclient.go

@@ -10,12 +10,10 @@ import (
 	"strings"
 
 	"github.com/cohere-ai/tokenizer"
+	"github.com/tmc/langchaingo/llms"
 )
 
-var (
-	ErrEmptyResponse = errors.New("empty response")
-	ErrModelNotFound = errors.New("model not found")
-)
+var ErrModelNotFound = errors.New("model not found")
 
 type Client struct {
 	token      string
@@ -111,8 +109,8 @@ func (c *Client) CreateGeneration(ctx context.Context, r *GenerationRequest) (*G
 		return nil, fmt.Errorf("create request: %w", err)
 	}
 
-	req.Header.Set("content-type", "application/json")
-	req.Header.Set("authorization", "bearer "+c.token)
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "bearer "+c.token)
 
 	res, err := c.httpClient.Do(req)
 	if err != nil {
@@ -129,7 +127,7 @@ func (c *Client) CreateGeneration(ctx context.Context, r *GenerationRequest) (*G
 		if strings.HasPrefix(response.Message, "model not found") {
 			return nil, ErrModelNotFound
 		}
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 
 	var generation Generation

llms/ernie/erniellm.go

@@ -11,10 +11,7 @@ import (
 	"github.com/tmc/langchaingo/llms/ernie/internal/ernieclient"
 )
 
-var (
-	ErrEmptyResponse = errors.New("no response")
-	ErrCodeResponse  = errors.New("has error code")
-)
+var ErrCodeResponse = errors.New("has error code")
 
 type LLM struct {
 	client           *ernieclient.Client

llms/ernie/internal/ernieclient/ernieclient.go

@@ -11,14 +11,15 @@ import (
 	"net/http"
 	"strings"
 	"time"
+
+	"github.com/tmc/langchaingo/llms"
 )
 
 var (
 	ErrNotSetAuth      = errors.New("both accessToken and apiKey secretKey are not set")
 	ErrCompletionCode  = errors.New("completion API returned unexpected status code")
 	ErrAccessTokenCode = errors.New("get access_token API returned unexpected status code")
 	ErrEmbeddingCode   = errors.New("embedding API returned unexpected status code")
-	ErrEmptyResponse   = errors.New("empty response")
 )
 
 // Client is a client for the ERNIE API.
@@ -285,7 +286,7 @@ func (c *Client) CreateChat(ctx context.Context, r *ChatRequest) (*ChatResponse,
 	}
 
 	if resp.Result == "" && resp.FunctionCall == nil {
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 
 	return resp, nil

llms/errors.go

@@ -0,0 +1,11 @@
+package llms
+
+import "errors"
+
+var (
+	// ErrEmptyResponse is returned when an LLM returns an empty response.
+	ErrEmptyResponse = errors.New("no response")
+	// ErrIncompleteEmbedding is returned when the length of an embedding
+	// request does not match the length of the returned embeddings array.
+	ErrIncompleteEmbedding = errors.New("not all input got embedded")
+)

llms/googleai/internal/palmclient/palmclient.go

@@ -21,10 +21,10 @@ var (
 )
 
 var defaultParameters = map[string]interface{}{ //nolint:gochecknoglobals
-	"temperature":     0.2, //nolint:gomnd
-	"maxOutputTokens": 256, //nolint:gomnd
-	"topP":            0.8, //nolint:gomnd
-	"topK":            40,  //nolint:gomnd
+	"temperature":     0.2, //nolint:all
+	"maxOutputTokens": 256, //nolint:all
+	"topP":            0.8, //nolint:all
+	"topK":            40,  //nolint:all
 }
 
 const (
@@ -65,9 +65,6 @@ func New(ctx context.Context, projectID, location string, opts ...option.ClientO
 	}, nil
 }
 
-// ErrEmptyResponse is returned when the OpenAI API returns an empty response.
-var ErrEmptyResponse = errors.New("empty response")
-
 // CompletionRequest is a request to create a completion.
 type CompletionRequest struct {
 	Prompts       []string `json:"prompts"`
@@ -290,7 +287,7 @@ func (c *PaLMClient) batchPredict(ctx context.Context, model string, prompts []s
 		return nil, err
 	}
 	if len(resp.GetPredictions()) == 0 {
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 	return resp.GetPredictions(), nil
 }
@@ -329,7 +326,7 @@ func (c *PaLMClient) chat(ctx context.Context, r *ChatRequest) ([]*structpb.Valu
 		return nil, err
 	}
 	if len(resp.GetPredictions()) == 0 {
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 	return resp.GetPredictions(), nil
 }

llms/googleai/option.go

@@ -25,18 +25,31 @@ type Options struct {
 	ClientOptions []option.ClientOption
 }
 
+const (
+	CloudProject          = ""
+	CloudLocation         = ""
+	DefaultModel          = "gemini-pro"
+	DefaultEmbeddingModel = "embedding-001"
+	DefaultCandidateCount = 1
+	DefaultMaxTokens      = 2048
+	DefaultTemperature    = 0.5
+	DefaultTopK           = 3
+	DefaultTopP           = 0.95
+	DefaultHarmThreshold  = HarmBlockOnlyHigh
+)
+
 func DefaultOptions() Options {
 	return Options{
-		CloudProject:          "",
-		CloudLocation:         "",
-		DefaultModel:          "gemini-pro",
-		DefaultEmbeddingModel: "embedding-001",
-		DefaultCandidateCount: 1,
-		DefaultMaxTokens:      2048,
-		DefaultTemperature:    0.5,
-		DefaultTopK:           3,
-		DefaultTopP:           0.95,
-		HarmThreshold:         HarmBlockOnlyHigh,
+		CloudProject:          CloudProject,
+		CloudLocation:         CloudLocation,
+		DefaultModel:          DefaultModel,
+		DefaultEmbeddingModel: DefaultEmbeddingModel,
+		DefaultCandidateCount: DefaultCandidateCount,
+		DefaultMaxTokens:      DefaultMaxTokens,
+		DefaultTemperature:    DefaultTemperature,
+		DefaultTopK:           DefaultTopK,
+		DefaultTopP:           DefaultTopP,
+		HarmThreshold:         DefaultHarmThreshold,
 	}
 }
 

llms/googleai/palm/palm_llm.go

@@ -13,7 +13,6 @@ import (
 )
 
 var (
-	ErrEmptyResponse            = errors.New("no response")
 	ErrMissingProjectID         = errors.New("missing the GCP Project ID, set it in the GOOGLE_CLOUD_PROJECT environment variable") //nolint:lll
 	ErrMissingLocation          = errors.New("missing the GCP Location, set it in the GOOGLE_CLOUD_LOCATION environment variable")  //nolint:lll
 	ErrUnexpectedResponseLength = errors.New("unexpected length of response")
@@ -85,7 +84,7 @@ func (o *LLM) CreateEmbedding(ctx context.Context, inputTexts []string) ([][]flo
 	}
 
 	if len(embeddings) == 0 {
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 	if len(inputTexts) != len(embeddings) {
 		return embeddings, ErrUnexpectedResponseLength

llms/huggingface/huggingfacellm.go

@@ -11,7 +11,6 @@ import (
 )
 
 var (
-	ErrEmptyResponse            = errors.New("empty response")
 	ErrMissingToken             = errors.New("missing the Hugging Face API token. Set it in the HUGGINGFACEHUB_API_TOKEN environment variable") //nolint:lll
 	ErrUnexpectedResponseLength = errors.New("unexpected length of response")
 )
@@ -115,7 +114,7 @@ func (o *LLM) CreateEmbedding(
 		return nil, err
 	}
 	if len(embeddings) == 0 {
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 	if len(inputTexts) != len(embeddings) {
 		return embeddings, ErrUnexpectedResponseLength

llms/huggingface/internal/huggingfaceclient/huggingfaceclient.go

@@ -4,13 +4,12 @@ import (
 	"context"
 	"errors"
 	"fmt"
-)
 
-var (
-	ErrInvalidToken  = errors.New("invalid token")
-	ErrEmptyResponse = errors.New("empty response")
+	"github.com/tmc/langchaingo/llms"
 )
 
+var ErrInvalidToken = errors.New("invalid token")
+
 type Client struct {
 	Token string
 	Model string
@@ -64,7 +63,7 @@ func (c *Client) RunInference(ctx context.Context, request *InferenceRequest) (*
 		return nil, fmt.Errorf("failed to run inference: %w", err)
 	}
 	if len(resp) == 0 {
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 	text := resp[0].Text
 	// TODO: Add response cleaning based on Model.
@@ -96,7 +95,7 @@ func (c *Client) CreateEmbedding(
 	}
 
 	if len(resp) == 0 {
-		return nil, ErrEmptyResponse
+		return nil, llms.ErrEmptyResponse
 	}
 
 	return resp, nil