Updated Readme

README.md

@@ -1,70 +1,150 @@
 ---
-license: apache-2.0
-language:
-- en
-- zh
-- es
-- fr
-- de
-- ja
-- ko
-- ru
-- ar
-metrics:
-- brier_score
-base_model:
-- Qwen/Qwen3-Embedding-0.6B
-new_version: Qwen/Qwen3-Embedding-0.6B
-pipeline_tag: sentence-similarity
-library_name: sentence-transformers
-tags:
-- code
-- semantic-search
-- multilingual
-- ttrpg
-- classification
-- embeddings
-sdk: gradio
+title: FFEmbeds
 emoji: 😻
-colorFrom: blue
-colorTo: pink
-thumbnail: >-
-  https://cdn-uploads.huggingface.co/production/uploads/69104657d1f603e82357d0fe/TRno0fZXYwAQ5GaqqBYM7.png
-short_description: Qwen3Embeds
+colorFrom: red
+colorTo: blue
+sdk: gradio
+sdk_version: 5.49.1
+license: apache-2.0
+short_description: Qwen 4b Embeddings for Games
 ---
+# 😻 FF Embeddings API
+
+**Powered by Qwen3-Embedding-4B** - Advanced 4B-parameter multilingual embedding model with matryoshka truncation
+
+
+## 🚀 Quick Start
+
+### Python
+
+```python
+import requests
+import numpy as np
+
+url = "https://YOUR_USERNAME-fairfate-embeddings.hf.space/api/predict"
+
+# Generate embeddings
+texts = [
+    "Storm King's Thunder - Epic D&D 5E giant-themed adventure",
+    "Curse of Strahd - Gothic horror campaign in Ravenloft"
+]
+
+response = requests.post(
+    url,
+    json={
+        "data": [texts, True, 1536],  # [texts, use_instruction, dimensions]
+        "fn_index": 0
+    }
+)
+
+result = response.json()
+embeddings = np.array(result["data"][0])
+
+# Calculate similarity
+def cosine_similarity(a, b):
+    return np.dot(a, b)  # Already normalized
+
+similarity = cosine_similarity(embeddings[0], embeddings[1])
+print(f"Similarity: {similarity:.3f}")  # ~0.65 (related but different)
+```
+
+### TypeScript/JavaScript
+
+```typescript
+interface EmbeddingResponse {
+  data: number[][][];
+}
+
+async function getEmbeddings(
+  texts: string[],
+  useInstruction: boolean = true,
+  dimensions: number = 1536
+): Promise<number[][]> {
+  const response = await fetch(
+    'https://YOUR_USERNAME-fairfate-embeddings.hf.space/api/predict',
+    {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        data: [texts, useInstruction, dimensions],
+        fn_index: 0
+      })
+    }
+  );
+
+  const result: EmbeddingResponse = await response.json();
+  return result.data[0];
+}
+
+// Usage
+const embeddings = await getEmbeddings([
+  "Pathfinder 2E Core Rulebook",
+  "D&D 5E Player's Handbook"
+]);
+
+// Calculate cosine similarity (already normalized)
+function cosineSimilarity(a: number[], b: number[]): number {
+  return a.reduce((sum, val, i) => sum + val * b[i], 0);
+}
+
+const similarity = cosineSimilarity(embeddings[0], embeddings[1]);
+console.log(`Similarity: ${similarity.toFixed(3)}`); // ~0.78 (both RPG rulebooks)
+```
+
+### cURL
+
+```bash
+curl -X POST \
+  https://YOUR_USERNAME-fairfate-embeddings.hf.space/api/predict \
+  -H "Content-Type: application/json" \
+  -d '{
+    "data": [
+      ["Cyberpunk detective noir campaign for Shadowrun"],
+      true,
+      1536
+    ],
+    "fn_index": 0
+  }'
+```
 
-# Tini Embeddings
+## 📖 API Reference
 
-**Powered by Qwen3-Embedding-0.6B** 
+### Endpoint
 
-Semantic embeddings service for tabletop RPG games
+```
+POST /api/predict
+```
 
-## Highlights
+### Request Format
 
-- **100+ Languages** - English, Spanish, French, German, Chinese, Japanese, Korean, Arabic, and more
-- **1024 Dimensions** - Flexible output (32-1024)
-- **32K Context** - Massive text support (vs 512 for most models)
-- **Fine-tuned** - tabletop RPG content / jargon
+```json
+{
+  "data": [
+    ["text1", "text2", ...],  // Array of texts to embed
+    true,                      // use_instruction (boolean)
+    1536                       // output_dimensions (32-3584, production: 1536)
+  ],
+  "fn_index": 0
+}
+```
 
-## Perfect For
+### Response Format
 
-- **Game Classification**: Auto-tag RPG products by genre, game system, themes
-- **Semantic Search**: In TTRPG context
-- **Recommendations**: "Games similar to this content"
-- **Duplicate Detection**: Find similar or duplicate listings
-- **Content Moderation**: Detect inappropriate content
-- **Multilingual Matching**: Multi-language game similarity
+```json
+{
+  "data": [
+    [
+      [0.123, -0.456, 0.789, ...],  // First embedding (1536 dims)
+      [-0.234, 0.567, -0.890, ...]  // Second embedding (1536 dims)
+    ]
+  ]
+}
+```
 
-## Model Information
+### Parameters
 
-| Property | Value |
-|----------|-------|
-| **Model** | [`Qwen/Qwen3-Embedding-0.6B`](https://huggingface.co/Qwen/Qwen3-Embedding-0.6B) |
-| **Dimensions** | 1024 (flexible: 32-1024) |
-| **Max Tokens** | 32,768 |
-| **Languages** | 100+ |
-| **Parameters** | 0.6B |
-| **Layers** | 28 |
-| **Normalization** | L2 normalized |
-| **Distance Metric** | Cosine similarity |
-| **MTEB Score** | 70.58 (multilingual) |
+| Parameter | Type | Default | Range | Description |
+|-----------|------|---------|-------|-------------|
+| `texts` | string[] | required | - | Texts to embed |
+| `use_instruction` | boolean | true | - | Add instruction prefix (recommended) |
+| `output_dimensions` | number | 1536 | 32-2560 | Output embedding size (production: 1536) |