tkmind/tkmind_go
1
0
Fork 0
Code Issues Pull Requests Actions 4 Packages Projects Releases Wiki Activity

Apps token limit (#7474)

Browse Source 
Co-authored-by: Douwe Osinga <douwe@squareup.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
This commit is contained in:
Douwe Osinga
2026-02-24 14:41:07 -05:00
committed by GitHub
parent 5e59b4dcb2
commit e7b15661cd
9 changed files with 52 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/goose/src/providers/formats/databricks.rs
+22 -24
View File
@@ -613,19 +613,19 @@ pub fn create_request(
let is_thinking_enabled = std::env::var("CLAUDE_THINKING_ENABLED").is_ok();
if is_claude_sonnet && is_thinking_enabled {
// Minimum budget_tokens is 1024
let budget_tokens = std::env::var("CLAUDE_THINKING_BUDGET")
.unwrap_or_else(|_| "16000".to_string())
.parse()
.unwrap_or(16000);
// Anthropic requires budget_tokens >= 1024
const DEFAULT_THINKING_BUDGET: i32 = 16000;
let budget_tokens: i32 = std::env::var("CLAUDE_THINKING_BUDGET")
.ok()
.and_then(|s| s.parse().ok())
.unwrap_or(DEFAULT_THINKING_BUDGET);
// For Claude models with thinking enabled, we need to add max_tokens + budget_tokens
// Default to 8192 (Claude max output) + budget if not specified
let max_completion_tokens = model_config.max_tokens.unwrap_or(8192);
payload.as_object_mut().unwrap().insert(
"max_tokens".to_string(),
json!(max_completion_tokens + budget_tokens),
);
// With thinking enabled, max_tokens must include both output and thinking budget
let max_tokens = model_config.max_output_tokens() + budget_tokens;
payload
.as_object_mut()
.unwrap()
.insert("max_tokens".to_string(), json!(max_tokens));
payload.as_object_mut().unwrap().insert(
"thinking".to_string(),
@@ -650,18 +650,16 @@ pub fn create_request(
}
}
// open ai reasoning models use max_completion_tokens instead of max_tokens
if let Some(tokens) = model_config.max_tokens {
let key = if is_openai_reasoning_model {
"max_completion_tokens"
} else {
"max_tokens"
};
payload
.as_object_mut()
.unwrap()
.insert(key.to_string(), json!(tokens));
}
// OpenAI reasoning models use max_completion_tokens instead of max_tokens
let key = if is_openai_reasoning_model {
"max_completion_tokens"
} else {
"max_tokens"
};
payload
.as_object_mut()
.unwrap()
.insert(key.to_string(), json!(model_config.max_output_tokens()));
}
// Apply cache control for Claude models to enable prompt caching