diff --git a/ee/app/assets/javascripts/ai/constants.js b/ee/app/assets/javascripts/ai/constants.js
index d433183dc90e1801cd0877144c9428f9f8529545..f1bb4faab273f835712f9e7bf9640f9c864cb1d9 100644
--- a/ee/app/assets/javascripts/ai/constants.js
+++ b/ee/app/assets/javascripts/ai/constants.js
@@ -49,8 +49,4 @@ export const FEEDBACK_OPTIONS = [
},
];
-const MODEL_MAX_TOKENS = 4096; // max for 'gpt-3.5-turbo' model as per https://platform.openai.com/docs/models/gpt-3-5
-export const MAX_RESPONSE_TOKENS = 300;
-export const TOKENS_THRESHOLD = MODEL_MAX_TOKENS * 0.85; // We account for 15% fault in tokens calculation
-
export const EXPLAIN_CODE_TRACKING_EVENT_NAME = 'explain_code_blob_viewer';
diff --git a/ee/app/assets/javascripts/ai/utils.js b/ee/app/assets/javascripts/ai/utils.js
index 3c5822bef8072839765168627851f5d48393ee4e..93a5d449ccc39f4052947609222e31364de8c4d9 100644
--- a/ee/app/assets/javascripts/ai/utils.js
+++ b/ee/app/assets/javascripts/ai/utils.js
@@ -1,14 +1,13 @@
import { findLastIndex } from 'lodash';
import { sprintf, __ } from '~/locale';
-import {
- TOO_LONG_ERROR_TYPE,
- i18n,
- MAX_RESPONSE_TOKENS,
- TOKENS_THRESHOLD,
- GENIE_CHAT_MODEL_ROLES,
-} from './constants';
+import { TOO_LONG_ERROR_TYPE, i18n, GENIE_CHAT_MODEL_ROLES } from './constants';
const areMessagesWithinLimit = (messages) => {
+ const MAX_RESPONSE_TOKENS = gon.ai?.chat?.max_response_token;
+ const TOKENS_THRESHOLD = gon.ai?.chat?.input_content_limit;
+
+ if (!MAX_RESPONSE_TOKENS || !TOKENS_THRESHOLD) return true; // delegate dealing with the prompt size to BE
+
// we use `utils.computeTokens()` below to make it easier to test and mock calls to computeTokens()
// eslint-disable-next-line no-use-before-define
return utils.computeTokens(messages) + MAX_RESPONSE_TOKENS < TOKENS_THRESHOLD;
diff --git a/ee/lib/ee/gitlab/gon_helper.rb b/ee/lib/ee/gitlab/gon_helper.rb
index 3e16fef268cbc423efbe7842149ca871c38edb09..633a8eebd61b8547e81201c04a3463ed3ab58eab 100644
--- a/ee/lib/ee/gitlab/gon_helper.rb
+++ b/ee/lib/ee/gitlab/gon_helper.rb
@@ -11,6 +11,16 @@ def add_gon_variables
gon.roadmap_epics_limit = 1000
+ if current_user && defined?(Llm)
+ ai_chat = {
+ total_model_token: ::Llm::ExplainCodeService::TOTAL_MODEL_TOKEN_LIMIT,
+ max_response_token: ::Llm::ExplainCodeService::MAX_RESPONSE_TOKENS,
+ input_content_limit: ::Llm::ExplainCodeService::INPUT_CONTENT_LIMIT
+ }
+
+ push_to_gon_attributes('ai', 'chat', ai_chat)
+ end
+
if ::Gitlab.com?
gon.subscriptions_url = ::Gitlab::Routing.url_helpers.subscription_portal_url
gon.subscriptions_legacy_sign_in_url = ::Gitlab::Routing.url_helpers.subscription_portal_legacy_sign_in_url
diff --git a/ee/spec/frontend/ai/utils_spec.js b/ee/spec/frontend/ai/utils_spec.js
index 80d0d0aa549e02542018efea62d29f4c4a7efff2..f6a0ef4235b2324139e3cacb46ff5fde6e1bb30d 100644
--- a/ee/spec/frontend/ai/utils_spec.js
+++ b/ee/spec/frontend/ai/utils_spec.js
@@ -1,27 +1,24 @@
import { utils } from 'ee/ai/utils';
-import {
- i18n,
- TOKENS_THRESHOLD,
- MAX_RESPONSE_TOKENS,
- GENIE_CHAT_MODEL_ROLES,
-} from 'ee/ai/constants';
+import { i18n, GENIE_CHAT_MODEL_ROLES } from 'ee/ai/constants';
import { sprintf } from '~/locale';
-jest.mock('ee/ai/constants', () => {
- // To simplify the things in testing, we override the constatants
- // to make the MAX_RESPONSE_TOKENS and TOKENS_THRESHOLD smaller
- // and easier to control
- const originalConstants = jest.requireActual('ee/ai/constants');
- return {
- ...originalConstants,
- TOKENS_THRESHOLD: 40, // 36 * 4 = 144 characters.
- MAX_RESPONSE_TOKENS: 4, // 4 * 4 = 16 characters.
- };
-});
-
+// To simplify the things in testing, we override the globals
+// to make the MAX_RESPONSE_TOKENS and TOKENS_THRESHOLD smaller
+// and easier to control
+const TOKENS_THRESHOLD = 40;
+const MAX_RESPONSE_TOKENS = 4;
const MAX_PROMPT_TOKENS = TOKENS_THRESHOLD - MAX_RESPONSE_TOKENS; // 36 tokens
describe('AI Utils', () => {
+ beforeEach(() => {
+ gon.ai = {
+ chat: {
+ max_response_token: MAX_RESPONSE_TOKENS,
+ input_content_limit: TOKENS_THRESHOLD,
+ },
+ };
+ });
+
describe('generateExplainCodePrompt', () => {
const filePath = 'fooPath';
const fileText = 'barText';
@@ -156,6 +153,23 @@ describe('AI Utils', () => {
i18n.TOO_LONG_ERROR_MESSAGE,
);
});
+
+ it.each`
+ max_response_token | input_content_limit
+ ${123} | ${undefined}
+ ${undefined} | ${123}
+ ${undefined} | ${undefined}
+ `(
+ 'drops no messages if token limitations are not available (delegates dealing with the prompt to BE)',
+ ({ max_response_token, input_content_limit }) => {
+ gon.ai = {
+ chat: { max_response_token, input_content_limit },
+ };
+
+ result = utils.generateChatPrompt(userPrompt, basePrompts);
+ expect(result).toEqual([...basePrompts, lastUserMessage]);
+ },
+ );
});
});
diff --git a/ee/spec/lib/ee/gitlab/gon_helper_spec.rb b/ee/spec/lib/ee/gitlab/gon_helper_spec.rb
index c11c9ebb9cc4a884485d4b6af67819054388db1d..4a5b4e538640408fc80c153a6564f29ad8bb4385 100644
--- a/ee/spec/lib/ee/gitlab/gon_helper_spec.rb
+++ b/ee/spec/lib/ee/gitlab/gon_helper_spec.rb
@@ -18,6 +18,8 @@ def current_user
before do
allow(helper).to receive(:gon).and_return(gon)
+ allow(helper).to receive(:push_to_gon_attributes).and_return(nil)
+ allow(helper).to receive(:current_user).and_return(create(:user))
end
it 'includes ee exclusive settings' do
@@ -26,6 +28,17 @@ def current_user
helper.add_gon_variables
end
+ it 'adds AI gon attributes' do
+ ai_chat = {
+ total_model_token: ::Llm::ExplainCodeService::TOTAL_MODEL_TOKEN_LIMIT,
+ max_response_token: ::Llm::ExplainCodeService::MAX_RESPONSE_TOKENS,
+ input_content_limit: ::Llm::ExplainCodeService::INPUT_CONTENT_LIMIT
+ }
+ helper.add_gon_variables
+
+ expect(helper).to have_received(:push_to_gon_attributes).with('ai', 'chat', ai_chat)
+ end
+
context 'when GitLab.com' do
before do
allow(Gitlab).to receive(:com?).and_return(true)