Add support for a smaller model for fast tasks

static/index.html

@@ -426,67 +426,93 @@
                             <button class="btn btn-sm btn-circle btn-ghost absolute right-2 top-2">✕</button>
                         </form>
                     </div>
-                    <h2 class="text-lg font-bold">Private generation settings</h2>
-                    <p class="py-4">Detail your private LLM credentials under to draft and generate solution using
-                        private
-                        LLM</p>
+                    <!--LLM api settings-->
                     <div class="space-y-4">
-                        <div>
-                            <label for="provider-url" class="block mb-2 text-sm font-medium">LLM provider URL</label>
-                            <input id="settings-provider-url" name="provider-url" class="input input-bordered w-full">
-                        </div>
+                        <div class="space-y-2">
+                            <h2 class="text-lg font-bold">LLM API access settings</h2>
+                            <p class="font-s">Detail your LLM credentials under to draft and generate solutions using
+                                the
+                                provided private LLM served over a OpenAI-like API</p>
+                            <div class="space-y-4">
+                                <div>
+                                    <label for="provider-url" class="block mb-2 text-sm font-medium">LLM provider URL
+                                        (without last leading /)</label>
+                                    <input id="settings-provider-url" name="provider-url"
+                                        class="input input-bordered w-full">
+                                </div>
 
-                        <div>
-                            <label for="provider-token" class="block mb-2 text-sm font-medium">LLM provider
-                                token</label>
-                            <input id="settings-provider-token" name="provider-token"
-                                class="input input-bordered w-full" type="password">
+                                <div>
+                                    <label for="provider-token" class="block mb-2 text-sm font-medium">LLM provider
+                                        token</label>
+                                    <input id="settings-provider-token" name="provider-token"
+                                        class="input input-bordered w-full" type="password">
+                                </div>
+                            </div>
                         </div>
-
-                        <div>
-                            <label for="provider-model" class="block mb-2 text-sm font-medium">Model</label>
+                        <!--LLM model selection for each task-->
+                        <div class="space-y-2">
+                            <h3 class="text-m font-bold">Model selection</h3>
+                            <p class="font-s">Select the models to use for solution evaluation, refinement and
+                                summarization.</p>
                             <button id="settings-fetch-models" class="btn btn-outline">Fetch models</button>
-                            <select id="settings-provider-model" name="provider-model" class="select">
-                            </select>
+                            <div>
+                                <label for="reasoning-model-selection" class="block mb-2 text-sm font-medium">Model for
+                                    reasoning tasks (evaluation, assessment)</label>
+                                <select id="settings-reasoning-model" name="reasoning-model-selection" class="select">
+                                </select>
+                            </div>
+                            <div>
+                                <label for="fast-model-selection" class="block mb-2 text-sm font-medium">Model for fast,
+                                    simple tasks (summarizing, structured output)</label>
+                                <select id="settings-fast-model" name="fast-model-selection" class="select">
+                                </select>
+                            </div>
                         </div>
 
-                        <div>
-                            <label for="assessment-rules" class="block mb-2 text-sm font-medium">Assessment
-                                rules</label>
-                            <textarea id="settings-assessment-rules" name="assessment-rules"
-                                class="textarea textarea-bordered w-full h-48"
-                                placeholder="Enter your rules here..."></textarea>
-                        </div>
+                        <!--Solution assessment settings-->
+                        <div class="space-y-2">
+                            <h3 class="text-m font-bold">Solution assessment settings</h3>
+                            <div>
+                                <label for="assessment-rules" class="block mb-2 text-sm font-medium">Assessment
+                                    rules</label>
+                                <textarea id="settings-assessment-rules" name="assessment-rules"
+                                    class="textarea textarea-bordered w-full h-48"
+                                    placeholder="Enter your rules here..."></textarea>
+                            </div>
 
-                        <div>
-                            <label for="portfolio-info" class="block mb-2 text-sm font-medium">Portfolio
-                                information</label>
-                            <textarea id="settings-portfolio" name="portfolio-info"
-                                class="textarea textarea-bordered w-full h-48"
-                                placeholder="Enter your portfolio info here..."></textarea>
-                        </div>
-                        <h2 class="text-lg font-bold">FTO analysis</h2>
-                        <p class="py-4">Configure the max numbers of topics explored in the FTO analysis.</p>
-                        <div class="w-full max-w-xs">
-                            <label for="settings-fto-topic-count">Number of topics to explore for FTO</label>
-                            <input id="settings-fto-topic-count" name="settings-fto-topic-count" type="range" min="1"
-                                max="10" value="4" class="range" step="1" />
-                            <!--c moche mais ca marche-->
-                            <div class="flex justify-between px-2.5 mt-2 text-xs">
-                                <span>|</span>
-                                <span></span>
-                                <span>|</span>
-                                <span></span>
-                                <span></span>
-                                <span>|</span>
+                            <div>
+                                <label for="portfolio-info" class="block mb-2 text-sm font-medium">Portfolio
+                                    information</label>
+                                <textarea id="settings-portfolio" name="portfolio-info"
+                                    class="textarea textarea-bordered w-full h-48"
+                                    placeholder="Enter your portfolio info here..."></textarea>
                             </div>
-                            <div class="flex justify-between px-2.5 mt-2 text-xs">
-                                <span>1</span>
-                                <span></span>
-                                <span>5</span>
-                                <span></span>
-                                <span></span>
-                                <span>10</span>
+                        </div>
+
+                        <div class="space-y-2">
+                            <h2 class="text-m font-bold">FTO analysis</h2>
+                            <p class="text-s">Configure the max numbers of topics explored in the FTO analysis.</p>
+                            <div class="w-full max-w-xs">
+                                <label for="settings-fto-topic-count">Number of topics to explore for FTO</label>
+                                <input id="settings-fto-topic-count" name="settings-fto-topic-count" type="range"
+                                    min="1" max="10" value="4" class="range" step="1" />
+                                <!--c moche mais ca marche-->
+                                <div class="flex justify-between px-2.5 mt-2 text-xs">
+                                    <span>|</span>
+                                    <span></span>
+                                    <span>|</span>
+                                    <span></span>
+                                    <span></span>
+                                    <span>|</span>
+                                </div>
+                                <div class="flex justify-between px-2.5 mt-2 text-xs">
+                                    <span>1</span>
+                                    <span></span>
+                                    <span>5</span>
+                                    <span></span>
+                                    <span></span>
+                                    <span>10</span>
+                                </div>
                             </div>
                         </div>
                         <p class="py-4">⚠ 'Save configuration' to keep the current configuration across sessions.</p>

static/js/app.js

@@ -1086,7 +1086,8 @@ document.getElementById('settings-fetch-models').addEventListener('click', _ =>
     const url = document.getElementById('settings-provider-url').value;
     const token = document.getElementById('settings-provider-token').value;
 
-    populateLLMModelSelect('settings-provider-model', url, token).catch(e => alert("Error while fetching models: " + e))
+    populateLLMModelSelect('settings-reasoning-model', url, token).catch(e => alert("Error while fetching models: " + e))
+    populateLLMModelSelect('settings-fast-model', url, token).catch(e => alert("Error while fetching models: " + e))
 });
 
 // ================== solution drafting events =============

static/js/gen.js

@@ -201,7 +201,7 @@ const StructuredAssessmentOutput = zod.object({
     insights: zod.array(zod.string()),
 });
 
-export async function assessSolution(providerUrl, modelName, apiKey, solution, assessment_rules, portfolio_info) {
+export async function assessSolution(providerUrl, reasoningModel, fastModel, apiKey, solution, assessment_rules, portfolio_info) {
     const template = await retrieveTemplate("assess");
 
     const assessment_template = formatTemplate(template, {
@@ -211,7 +211,7 @@ export async function assessSolution(providerUrl, modelName, apiKey, solution, a
         solution_description: solution.solution_description,
     });
 
-    const assessment_full = await generateCompletion(providerUrl, modelName, apiKey, [
+    const assessment_full = await generateCompletion(providerUrl, reasoningModel, apiKey, [
         { role: "user", content: assessment_template }
     ]);
 
@@ -221,7 +221,7 @@ export async function assessSolution(providerUrl, modelName, apiKey, solution, a
         "response_schema": zod.toJSONSchema(StructuredAssessmentOutput)
     })
 
-    const extracted_info = await generateStructuredCompletion(providerUrl, modelName, apiKey, [{ role: "user", content: structured_filled_template }], StructuredAssessmentOutput);
+    const extracted_info = await generateStructuredCompletion(providerUrl, fastModel, apiKey, [{ role: "user", content: structured_filled_template }], StructuredAssessmentOutput);
 
     return { assessment_full, extracted_info };
 }
@@ -258,7 +258,7 @@ const FTOAnalysisTopicsSchema = zod.object({
 /**
  * Extract the topics to search for FTO
  */
-async function getFtoAnalysisTopics(providerUrl, modelName, apiKey, idea, count) {
+async function getFtoAnalysisTopics(providerUrl, reasoningModel, apiKey, idea, count) {
     const template = await retrieveTemplate("fto_topics");
 
     const structured_template = formatTemplate(template, {
@@ -268,7 +268,7 @@ async function getFtoAnalysisTopics(providerUrl, modelName, apiKey, idea, count)
         "max_topic_count": count
     });
 
-    const topics = await generateStructuredCompletion(providerUrl, modelName, apiKey, [{ role: "user", content: structured_template }], FTOAnalysisTopicsSchema);
+    const topics = await generateStructuredCompletion(providerUrl, reasoningModel, apiKey, [{ role: "user", content: structured_template }], FTOAnalysisTopicsSchema);
 
     return topics;
 }
@@ -276,7 +276,7 @@ async function getFtoAnalysisTopics(providerUrl, modelName, apiKey, idea, count)
 /*
  * Assess the infringement of the idea wrt 
 */
-async function assessFTOReport(providerUrl, modelName, apiKey, solution, fto_report, portfolio_info) {
+async function assessFTOReport(providerUrl, reasoningModel, fastModel, apiKey, solution, fto_report, portfolio_info) {
     const template = await retrieveTemplate("fto_assess");
 
     const assessment_template = formatTemplate(template, {
@@ -288,7 +288,7 @@ async function assessFTOReport(providerUrl, modelName, apiKey, solution, fto_rep
 
     console.log("FTO Length: " + assessment_template.length);
 
-    const assessment_full = await generateCompletion(providerUrl, modelName, apiKey, [
+    const assessment_full = await generateCompletion(providerUrl, reasoningModel, apiKey, [
         { role: "user", content: assessment_template }
     ]);
 
@@ -298,18 +298,18 @@ async function assessFTOReport(providerUrl, modelName, apiKey, solution, fto_rep
         "response_schema": zod.toJSONSchema(StructuredAssessmentOutput)
     })
 
-    const extracted_info = await generateStructuredCompletion(providerUrl, modelName, apiKey, [{ role: "user", content: structured_filled_template }], StructuredAssessmentOutput);
+    const extracted_info = await generateStructuredCompletion(providerUrl, fastModel, apiKey, [{ role: "user", content: structured_filled_template }], StructuredAssessmentOutput);
 
     return { assessment_full, extracted_info };
 }
 
-export async function runFTOAnalysis(providerUrl, providerModel, apiKey, solution, portfolio_info, ftoTopicCount) {
-    const fto_topics = await getFtoAnalysisTopics(providerUrl, providerModel, apiKey, solution, ftoTopicCount);
+export async function runFTOAnalysis(providerUrl, reasoningModel, fastModel, apiKey, solution, portfolio_info, ftoTopicCount) {
+    const fto_topics = await getFtoAnalysisTopics(providerUrl, reasoningModel, apiKey, solution, ftoTopicCount);
     console.log(fto_topics);
 
     const fto_report = await performDeepSearch(fto_topics.topics);
 
-    const assess_results = await assessFTOReport(providerUrl, providerModel, apiKey, solution, fto_report, portfolio_info);
+    const assess_results = await assessFTOReport(providerUrl, reasoningModel, fastModel, apiKey, solution, fto_report, portfolio_info);
     console.log(assess_results.extracted_info);
 
     return {

static/js/persistence.js

@@ -1,5 +1,5 @@
 // clées utilisées pour la generation avec private compute
-const CONFIG_KEYS = ["providerUrl", "providerToken", "providerModel", "assessmentRules", "businessPortfolio", "ftoTopicCount"];
+const CONFIG_KEYS = ["providerUrl", "providerToken", "reasoningModel", "fastModel", "assessmentRules", "businessPortfolio", "ftoTopicCount"];
 // clée pour marquer si la config est remplie.
 const POPULATED_CONFIG_KEY = "_config_populated";
 

static/js/ui.js

@@ -356,21 +356,22 @@ export function debounceAutoCategoryCount(state) {
 export function getConfigFields() {
     const providerUrl = document.getElementById('settings-provider-url').value;
     const providerToken = document.getElementById('settings-provider-token').value;
-    const providerModel = document.getElementById('settings-provider-model').value;
+    const reasoningModel = document.getElementById('settings-reasoning-model').value;
+    const fastModel = document.getElementById('settings-fast-model').value;
     const assessmentRules = document.getElementById('settings-assessment-rules').value;
     const businessPortfolio = document.getElementById('settings-portfolio').value;
     const ftoTopicCount = document.getElementById('settings-fto-topic-count').value;
 
-    return { providerUrl, providerToken, providerModel, assessmentRules, businessPortfolio, ftoTopicCount };
+    return { providerUrl, providerToken, reasoningModel, fastModel, assessmentRules, businessPortfolio, ftoTopicCount };
 }
 
 /**
 *  Vérifie si les paramètres sont bien renseignés pour utiliser la génération privée.
 */
 export function checkPrivateLLMInfoAvailable() {
-    const { providerUrl, providerToken, providerModel, assessmentRules, businessPortfolio } = getConfigFields();
+    const { providerUrl, providerToken, reasoningModel, fastModel, assessmentRules, businessPortfolio } = getConfigFields();
     const isEmpty = (str) => (!str?.length);
-    return !isEmpty(providerUrl) && !isEmpty(providerToken) && !isEmpty(assessmentRules) && !isEmpty(businessPortfolio) && !isEmpty(providerModel);
+    return !isEmpty(providerUrl) && !isEmpty(providerToken) && !isEmpty(assessmentRules) && !isEmpty(businessPortfolio) && !isEmpty(reasoningModel) && !isEmpty(fastModel);
     // return true;
 }
 
@@ -433,7 +434,8 @@ export function handleLoadConfigFields() {
 
     const providerUrl = document.getElementById('settings-provider-url');
     const providerToken = document.getElementById('settings-provider-token');
-    const providerModel = document.getElementById('settings-provider-model');
+    const reasoningModel = document.getElementById('settings-reasoning-model');
+    const fastModel = document.getElementById('settings-fast-model');
     const assessmentRules = document.getElementById('settings-assessment-rules');
     const businessPortfolio = document.getElementById('settings-portfolio');
     const ftoTopicCount = document.getElementById('settings-fto-topic-count');
@@ -443,12 +445,19 @@ export function handleLoadConfigFields() {
     assessmentRules.value = configuration.assessmentRules;
     businessPortfolio.value = configuration.businessPortfolio;
     ftoTopicCount.value = configuration.ftoTopicCount;
+    fastModel.value = configuration.fastModel;
 
     // on doit d'abord recup les modeles avant de set la valeur
-    populateLLMModelSelect('settings-provider-model', configuration.providerUrl, configuration.providerToken).then(() => {
-        providerModel.value = configuration.providerModel;
+    populateLLMModelSelect('settings-reasoning-model', configuration.providerUrl, configuration.providerToken).then(() => {
+        reasoningModel.value = configuration.reasoningModel;
     }).catch(e => {
-        alert("Failed to set LLM model in model selector. Model may not be available anymore, check model in LLM settings.");
+        alert("Failed to set LLM model for reasoning in the model selector. Model may not be available anymore, check LLM settings.");
+    })
+
+    populateLLMModelSelect('settings-fast-model', configuration.providerUrl, configuration.providerToken).then(() => {
+        fastModel.value = configuration.fastModel;
+    }).catch(e => {
+        alert("Failed to set fast LLM model in the model selector. Model may not be available anymore, check LLM settings.");
     })
 }
 
@@ -499,10 +508,10 @@ export function moveSolutionToDrafts(solution) {
 
     switchTab('draft-tab');
 
-    const { providerUrl, providerToken, providerModel, assessmentRules, businessPortfolio } = getConfigFields();
+    const { providerUrl, providerToken, reasoningModel, fastModel, assessmentRules, businessPortfolio } = getConfigFields();
 
     showLoadingOverlay("Assessing solution ....");
-    assessSolution(providerUrl, providerModel, providerToken, solution, assessmentRules, businessPortfolio).then(response => {
+    assessSolution(providerUrl, reasoningModel, fastModel, providerToken, solution, assessmentRules, businessPortfolio).then(response => {
         // reset the state of the draft history
         draftHistory = [];
         draftCurrentIndex = -1;
@@ -645,14 +654,14 @@ export function handleDraftRefine() {
     }
     // ---
 
-    const { providerUrl, providerToken, providerModel, assessmentRules, businessPortfolio } = getConfigFields();
+    const { providerUrl, providerToken, reasoningModel, fastModel, assessmentRules, businessPortfolio } = getConfigFields();
 
     showLoadingOverlay('Refining and assessing ....')
 
-    refineSolution(providerUrl, providerModel, providerToken, currentState.solution, selectedInsights, userInsightsText, assessmentRules, businessPortfolio)
+    refineSolution(providerUrl, reasoningModel, providerToken, currentState.solution, selectedInsights, userInsightsText, assessmentRules, businessPortfolio)
         .then(newSolution => {
             const refinedSolution = newSolution;
-            return assessSolution(providerUrl, providerModel, providerToken, newSolution, assessmentRules, businessPortfolio)
+            return assessSolution(providerUrl, reasoningModel, fastModel, providerToken, newSolution, assessmentRules, businessPortfolio)
                 .then(assessedResult => {
                     return { refinedSolution, assessedResult };
                 });
@@ -692,14 +701,14 @@ function jumpToDraft(index) {
 }
 
 export function handleFTOAnalysis() {
-    const { providerUrl, providerToken, providerModel, businessPortfolio, ftoTopicCount } = getConfigFields();
+    const { providerUrl, providerToken, reasoningModel, fastModel, businessPortfolio, ftoTopicCount } = getConfigFields();
     const currentState = draftHistory[draftCurrentIndex];
 
     console.log("Launching FTO analysis");
 
     showLoadingOverlay("Running FTO analysis... This may take a while");
 
-    runFTOAnalysis(providerUrl, providerModel, providerToken, currentState.solution, businessPortfolio, ftoTopicCount)
+    runFTOAnalysis(providerUrl, reasoningModel, fastModel, providerToken, currentState.solution, businessPortfolio, ftoTopicCount)
         .then(result => {
             // map from a list of insights to a selectable list of insights
             const newInsights = result.extracted_info.insights.map((e, idx, __) => ({ id: idx, text: e, checked: false }));