fix(tools): route quick/pro tiers to Haiku/Sonnet on Bedrock
Tier engine strings (claude_haiku/claude_sonnet) were stripped back to azure_mini by per-method whitelists, so both tiers ran gpt-4o-mini and Pro charged 2x for the same model. Add a shared DbnBedrockModelRouter:: deploymentForEngine() helper and route the cloud path through it across summarize, ask, barnevernet, discrepancy, deep-research, and korrespond. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
@@ -212,7 +212,7 @@ final class DbnLegalToolsService
|
||||
|
||||
public function ask(string $question, string $language = 'en', string $engine = 'azure_mini', ?string $persona = null): array
|
||||
{
|
||||
$engine = in_array($engine, ['azure_mini', 'azure_full'], true) ? $engine : 'azure_mini';
|
||||
$engine = in_array($engine, ['azure_mini', 'azure_full', 'claude_haiku', 'claude_sonnet'], true) ? $engine : 'azure_mini';
|
||||
$client = dbnToolsRequireClient();
|
||||
$personaResolved = dbnToolsResolvePersona((int)$client['id'], $persona);
|
||||
$search = $this->search($question, $language, 7, 'disabled', null, 'both', $personaResolved['slug']);
|
||||
@@ -1231,7 +1231,7 @@ PROMPT;
|
||||
error_log('[dbn-persona] gateway init failed for model ' . $model . ': ' . $e->getMessage());
|
||||
}
|
||||
}
|
||||
return [$this->azure, ($engine === 'azure_full') ? 'gpt-4o' : 'gpt-4o-mini'];
|
||||
return [$this->azure, DbnBedrockModelRouter::deploymentForEngine($engine, $this->azure instanceof DbnBedrockGateway)];
|
||||
}
|
||||
|
||||
/**
|
||||
@@ -2060,7 +2060,7 @@ PROMPT;
|
||||
string $depth = 'standard'
|
||||
): array {
|
||||
$text = $this->requirePasteText($text);
|
||||
$engine = in_array($engine, ['azure_mini', 'azure_full', 'gpu'], true) ? $engine : 'azure_mini';
|
||||
$engine = in_array($engine, ['azure_mini', 'azure_full', 'gpu', 'claude_haiku', 'claude_sonnet'], true) ? $engine : 'azure_mini';
|
||||
|
||||
$locale = dbnToolsLanguageName($language);
|
||||
|
||||
@@ -2114,7 +2114,7 @@ PROMPT;
|
||||
['role' => 'system', 'content' => $system],
|
||||
['role' => 'user', 'content' => $prompt],
|
||||
];
|
||||
$maxTok = ($engine === 'azure_full') ? 8000 : 4000;
|
||||
$maxTok = in_array($engine, ['azure_full', 'claude_sonnet'], true) ? 8000 : 4000;
|
||||
$chatOpts = ['json' => true, 'temperature' => 0.1, 'max_tokens' => $maxTok, 'timeout' => 120];
|
||||
|
||||
$deployLabel = $this->azure->chatDeployment();
|
||||
@@ -2122,12 +2122,10 @@ PROMPT;
|
||||
if ($engine === 'gpu') {
|
||||
$response = $this->callGpuLlm($messages, $chatOpts);
|
||||
$deployLabel = 'GPU (local)';
|
||||
} elseif ($engine === 'azure_full') {
|
||||
$response = $this->azure->withDeployment('gpt-4o')->chat($messages, $chatOpts);
|
||||
$deployLabel = 'gpt-4o';
|
||||
} else {
|
||||
$response = $this->azure->withDeployment('gpt-4o-mini')->chat($messages, $chatOpts);
|
||||
$deployLabel = 'gpt-4o-mini';
|
||||
$deploy = DbnBedrockModelRouter::deploymentForEngine($engine, $this->azure instanceof DbnBedrockGateway);
|
||||
$response = $this->azure->withDeployment($deploy)->chat($messages, $chatOpts);
|
||||
$deployLabel = $deploy;
|
||||
}
|
||||
} catch (Throwable $e) {
|
||||
dbnToolsAbort('LLM request failed: ' . $e->getMessage(), 502, 'llm_error');
|
||||
|
||||
Reference in New Issue
Block a user