oak-team/oak-gpui
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

ollama: Properly format tool calls fed back to the model (#34750)

Browse Source 
Fix an issue that resulted in Ollama models not being able to not being
able to access the input of the commands they executed (only being able
to access the result).

This properly return the function history as shown in
https://github.com/ollama/ollama/blob/main/docs/api.md#chat-request-with-history-with-tools

Previously, function input where not returned and result where returned
as a "user" role.

Release Notes:

- ollama: Improved format when returning tool results to the models
This commit is contained in:
marius851000
2025-09-08 04:26:01 +00:00
committed by GitHub
parent 69bdef38ec
commit 9450bcad25
2 changed files with 80 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/language_models/src/provider/ollama.rs
+76 -47
View File
@@ -11,8 +11,8 @@ use language_model::{
LanguageModelToolUseId, MessageContent, RateLimiter, Role, StopReason, TokenUsage,
};
use ollama::{
ChatMessage, ChatOptions, ChatRequest, ChatResponseDelta, KeepAlive, OllamaFunctionTool,
OllamaToolCall, get_models, show_model, stream_chat_completion,
ChatMessage, ChatOptions, ChatRequest, ChatResponseDelta, KeepAlive, OllamaFunctionCall,
OllamaFunctionTool, OllamaToolCall, get_models, show_model, stream_chat_completion,
};
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
@@ -282,59 +282,85 @@ impl OllamaLanguageModel {
fn to_ollama_request(&self, request: LanguageModelRequest) -> ChatRequest {
let supports_vision = self.model.supports_vision.unwrap_or(false);
ChatRequest {
model: self.model.name.clone(),
messages: request
.messages
.into_iter()
.map(|msg| {
let images = if supports_vision {
msg.content
.iter()
.filter_map(|content| match content {
MessageContent::Image(image) => Some(image.source.to_string()),
_ => None,
})
.collect::<Vec<String>>()
} else {
vec![]
};
let mut messages = Vec::with_capacity(request.messages.len());
match msg.role {
Role::User => ChatMessage::User {
for mut msg in request.messages.into_iter() {
let images = if supports_vision {
msg.content
.iter()
.filter_map(|content| match content {
MessageContent::Image(image) => Some(image.source.to_string()),
_ => None,
})
.collect::<Vec<String>>()
} else {
vec![]
};
match msg.role {
Role::User => {
for tool_result in msg
.content
.extract_if(.., |x| matches!(x, MessageContent::ToolResult(..)))
{
match tool_result {
MessageContent::ToolResult(tool_result) => {
messages.push(ChatMessage::Tool {
tool_name: tool_result.tool_name.to_string(),
content: tool_result.content.to_str().unwrap_or("").to_string(),
})
}
_ => unreachable!("Only tool result should be extracted"),
}
}
if !msg.content.is_empty() {
messages.push(ChatMessage::User {
content: msg.string_contents(),
images: if images.is_empty() {
None
} else {
Some(images)
},
},
Role::Assistant => {
let content = msg.string_contents();
let thinking =
msg.content.into_iter().find_map(|content| match content {
MessageContent::Thinking { text, .. } if !text.is_empty() => {
Some(text)
}
_ => None,
});
ChatMessage::Assistant {
content,
tool_calls: None,
images: if images.is_empty() {
None
} else {
Some(images)
},
thinking,
}
}
Role::System => ChatMessage::System {
content: msg.string_contents(),
},
})
}
})
.collect(),
}
Role::Assistant => {
let content = msg.string_contents();
let mut thinking = None;
let mut tool_calls = Vec::new();
for content in msg.content.into_iter() {
match content {
MessageContent::Thinking { text, .. } if !text.is_empty() => {
thinking = Some(text)
}
MessageContent::ToolUse(tool_use) => {
tool_calls.push(OllamaToolCall::Function(OllamaFunctionCall {
name: tool_use.name.to_string(),
arguments: tool_use.input,
}));
}
_ => (),
}
}
messages.push(ChatMessage::Assistant {
content,
tool_calls: Some(tool_calls),
images: if images.is_empty() {
None
} else {
Some(images)
},
thinking,
})
}
Role::System => messages.push(ChatMessage::System {
content: msg.string_contents(),
}),
}
}
ChatRequest {
model: self.model.name.clone(),
messages,
keep_alive: self.model.keep_alive.clone().unwrap_or_default(),
stream: true,
options: Some(ChatOptions {
@@ -483,6 +509,9 @@ fn map_to_language_model_completion_events(
ChatMessage::System { content } => {
events.push(Ok(LanguageModelCompletionEvent::Text(content)));
}
ChatMessage::Tool { content, .. } => {
events.push(Ok(LanguageModelCompletionEvent::Text(content)));
}
ChatMessage::Assistant {
content,
tool_calls,
crates/ollama/src/ollama.rs
+4
View File
@@ -117,6 +117,10 @@ pub enum ChatMessage {
System {
content: String,
},
Tool {
tool_name: String,
content: String,
},
}
#[derive(Serialize, Deserialize, Debug)]