Yehowshua/ZIm
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Accept wrapped text content from LLM providers (#31048)

Browse source 
Some providers sometimes send `{ "type": "text", "text": ... }` instead
of just the text as a string. Now we accept those instead of erroring.

Release Notes:

- N/A
This commit is contained in:
Richard Feldman 2025-05-20 16:50:02 -04:00 • committed by GitHub
parent 89700c3682
commit 4bb04cef9d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
9 changed files with 72 additions and 26 deletions
Download patch file Download diff file Expand all files Collapse all files

17
crates/language_models/src/provider/anthropic.rs
View file

@ -19,7 +19,7 @@ use language_model::{
LanguageModelCompletionError, LanguageModelId, LanguageModelKnownError, LanguageModelName,
LanguageModelProvider, LanguageModelProviderId, LanguageModelProviderName,
LanguageModelProviderState, LanguageModelRequest, LanguageModelToolChoice,
LanguageModelToolResultContent, MessageContent, RateLimiter, Role,
LanguageModelToolResultContent, MessageContent, RateLimiter, Role, WrappedTextContent,
};
use language_model::{LanguageModelCompletionEvent, LanguageModelToolUse, StopReason};
use schemars::JsonSchema;
@ -350,8 +350,12 @@ pub fn count_anthropic_tokens(
// TODO: Estimate token usage from tool uses.
}
MessageContent::ToolResult(tool_result) => match &tool_result.content {
LanguageModelToolResultContent::Text(txt) => {
string_contents.push_str(txt);
LanguageModelToolResultContent::Text(text)
| LanguageModelToolResultContent::WrappedText(WrappedTextContent {
text,
..
}) => {
string_contents.push_str(text);
}
LanguageModelToolResultContent::Image(image) => {
tokens_from_images += image.estimate_tokens();
@ -588,9 +592,10 @@ pub fn into_anthropic(
tool_use_id: tool_result.tool_use_id.to_string(),
is_error: tool_result.is_error,
content: match tool_result.content {
LanguageModelToolResultContent::Text(text) => {
ToolResultContent::Plain(text.to_string())
}
LanguageModelToolResultContent::Text(text)
| LanguageModelToolResultContent::WrappedText(
WrappedTextContent { text, .. },
) => ToolResultContent::Plain(text.to_string()),
LanguageModelToolResultContent::Image(image) => {
ToolResultContent::Multipart(vec![ToolResultPart::Image {
source: anthropic::ImageSource {

11
crates/language_models/src/provider/bedrock.rs
View file

@ -37,7 +37,7 @@ use language_model::{
LanguageModelProvider, LanguageModelProviderId, LanguageModelProviderName,
LanguageModelProviderState, LanguageModelRequest, LanguageModelToolChoice,
LanguageModelToolResultContent, LanguageModelToolUse, MessageContent, RateLimiter, Role,
TokenUsage,
TokenUsage, WrappedTextContent,
};
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
@ -641,7 +641,8 @@ pub fn into_bedrock(
BedrockToolResultBlock::builder()
.tool_use_id(tool_result.tool_use_id.to_string())
.content(match tool_result.content {
LanguageModelToolResultContent::Text(text) => {
LanguageModelToolResultContent::Text(text)
| LanguageModelToolResultContent::WrappedText(WrappedTextContent { text, .. }) => {
BedrockToolResultContentBlock::Text(text.to_string())
}
LanguageModelToolResultContent::Image(_) => {
@ -776,7 +777,11 @@ pub fn get_bedrock_tokens(
// TODO: Estimate token usage from tool uses.
}
MessageContent::ToolResult(tool_result) => match tool_result.content {
LanguageModelToolResultContent::Text(text) => {
LanguageModelToolResultContent::Text(text)
| LanguageModelToolResultContent::WrappedText(WrappedTextContent {
text,
..
}) => {
string_contents.push_str(&text);
}
LanguageModelToolResultContent::Image(image) => {

8
crates/language_models/src/provider/copilot_chat.rs
View file

@ -23,7 +23,7 @@ use language_model::{
LanguageModelProviderName, LanguageModelProviderState, LanguageModelRequest,
LanguageModelRequestMessage, LanguageModelToolChoice, LanguageModelToolResultContent,
LanguageModelToolSchemaFormat, LanguageModelToolUse, MessageContent, RateLimiter, Role,
StopReason,
StopReason, WrappedTextContent,
};
use settings::SettingsStore;
use std::time::Duration;
@ -455,7 +455,11 @@ fn into_copilot_chat(
for content in &message.content {
if let MessageContent::ToolResult(tool_result) = content {
let content = match &tool_result.content {
LanguageModelToolResultContent::Text(text) => text.to_string().into(),
LanguageModelToolResultContent::Text(text)
| LanguageModelToolResultContent::WrappedText(WrappedTextContent {
text,
..
}) => text.to_string().into(),
LanguageModelToolResultContent::Image(image) => {
if model.supports_vision() {
ChatMessageContent::Multipart(vec![ChatMessagePart::Image {

7
crates/language_models/src/provider/google.rs
View file

@ -426,14 +426,17 @@ pub fn into_google(
}
language_model::MessageContent::ToolResult(tool_result) => {
match tool_result.content {
language_model::LanguageModelToolResultContent::Text(txt) => {
language_model::LanguageModelToolResultContent::Text(text)
| language_model::LanguageModelToolResultContent::WrappedText(
language_model::WrappedTextContent { text, .. },
) => {
vec![Part::FunctionResponsePart(
google_ai::FunctionResponsePart {
function_response: google_ai::FunctionResponse {
name: tool_result.tool_name.to_string(),
// The API expects a valid JSON object
response: serde_json::json!({
"output": txt
"output": text
}),
},
},

8
crates/language_models/src/provider/mistral.rs
View file

@ -13,7 +13,7 @@ use language_model::{
LanguageModelId, LanguageModelName, LanguageModelProvider, LanguageModelProviderId,
LanguageModelProviderName, LanguageModelProviderState, LanguageModelRequest,
LanguageModelToolChoice, LanguageModelToolResultContent, LanguageModelToolUse, MessageContent,
RateLimiter, Role, StopReason,
RateLimiter, Role, StopReason, WrappedTextContent,
};
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
@ -428,7 +428,11 @@ pub fn into_mistral(
}
MessageContent::ToolResult(tool_result) => {
let content = match &tool_result.content {
LanguageModelToolResultContent::Text(text) => text.to_string(),
LanguageModelToolResultContent::Text(text)
| LanguageModelToolResultContent::WrappedText(WrappedTextContent {
text,
..
}) => text.to_string(),
LanguageModelToolResultContent::Image(_) => {
// TODO: Mistral image support
"[Tool responded with an image, but Zed doesn't support these in Mistral models yet]".to_string()

8
crates/language_models/src/provider/open_ai.rs
View file

@ -13,7 +13,7 @@ use language_model::{
LanguageModelId, LanguageModelName, LanguageModelProvider, LanguageModelProviderId,
LanguageModelProviderName, LanguageModelProviderState, LanguageModelRequest,
LanguageModelToolChoice, LanguageModelToolResultContent, LanguageModelToolUse, MessageContent,
RateLimiter, Role, StopReason,
RateLimiter, Role, StopReason, WrappedTextContent,
};
use open_ai::{ImageUrl, Model, ResponseStreamEvent, stream_completion};
use schemars::JsonSchema;
@ -407,7 +407,11 @@ pub fn into_open_ai(
}
MessageContent::ToolResult(tool_result) => {
let content = match &tool_result.content {
LanguageModelToolResultContent::Text(text) => {
LanguageModelToolResultContent::Text(text)
| LanguageModelToolResultContent::WrappedText(WrappedTextContent {
text,
..
}) => {
vec![open_ai::MessagePart::Text {
text: text.to_string(),
}]