Update README.md

Very first sketch of Organizational Card.

For the sinequa presentation I use the one from the recent press release. Feel free to modify it.



@basilevc


@skirres


@claeyzre

README.md

@@ -7,4 +7,77 @@ sdk: static
 pinned: false
 ---
 
-Edit this `README.md` markdown file to author your organization card.
+Sinequa Hugging Face homepage
+
+# About Sinequa
+
+Sinequa transforms how work gets done. Sinequa’s Assistants augment your company by augmenting employees with a knowledgeable, 
+accurate, secure work partner so they are more effective, more informed, more productive, and less stressed. Best of all, 
+Sinequa Assistants streamline workflows and automatically navigate the chaotic enterprise information landscape, so that employees 
+can skip the grind and focus on doing the kind of work that makes the most impact. Sinequa’s Assistants achieve this by combining 
+the power of comprehensive enterprise search with the ease of generative AI in a configurable and easily managed Assistant framework, 
+for an accurate, traceable, and fully secure conversational experience. Deploy an out-of-the-box Assistant or configure a tailored 
+experience and specialized workflow to augment your people and your company. For more information, visit www.sinequa.com.
+
+# Neural Search models
+
+Sinequa Search relies on a technology called Neural Search. Neural Search is an hybrid search solution based on both Key Word Search and Vector Searched.
+This search workflow implies thre type of models for which we deliver various version here.
+
+The two collections below bring together the recommended model combinations for English only and multilingual context.
+
+## Vectorizer
+
+Vectorizers are models which produce an embedding vector given a passage or a query. The passage vectors are stored in our vector index and the 
+query vector is used at query time to look up relevant passages in the index.
+
+Here is an overview of the model we deliver publicly here.
+
+Here’s the markdown table without formatting:  
+
+| Model                          | Languages                   | Relevance | Inference Time | GPU Memory |
+|--------------------------------|-----------------------------|-----------|----------------|------------|
+| vectorizer-v1-S-en             | en                          | 0.456     | 52 ms          | 330 MiB    |
+| vectorizer-v1-S-multilingual   | de, en, es, fr             | 0.448     | 51 ms          | 580 MiB    |
+| vectorizer.vanilla             | en                          | 0.639     | 53 ms          | 330 MiB    |
+| vectorizer.raspberry           | de, en, es, fr, it, ja, nl, pt, zs | 0.613     | 52 ms          | 610 MiB    |
+| vectorizer.hazelnut            | de, en, es, fr, it, ja, nl, pt, zs, pl | 0.590     | 52 ms          | 610 MiB    |
+
+## Passage Ranker
+
+Passage Rankers are models which produce a relevance score given a query-passage pair and is used to order search results coming from Key Word and Vector search.
+
+Here is an overview of the model we deliver publicly here.
+
+Here’s the table in markdown format:  
+
+| Model                           | Languages                   | Relevance | Inference Time | GPU Memory |
+|---------------------------------|-----------------------------|-----------|----------------|------------|
+| passage-ranker-v1-XS-en         | en                          | 0.438     | 20 ms          | 170 MiB    |
+| passage-ranker-v1-XS-multilingual | de, en, es, fr             | 0.453     | 21 ms          | 300 MiB    |
+| passage-ranker-v1-L-en          | en                          | 0.466     | 356 ms         | 1060 MiB   |
+| passage-ranker-v1-L-multilingual | de, en, es, fr             | 0.471     | 357 ms         | 1130 MiB   |
+| passage-ranker.chocolate        | en                          | 0.484     | 64 ms          | 550 MiB    |
+| passage-ranker.strawberry       | de, en, es, fr, it, ja, nl, pt, zs | 0.451     | 63 ms          | 1060 MiB   |
+| passage-ranker.mango            | de, en, es, fr, it, ja, nl, pt, zs | 0.480     | 358 ms         | 1070 MiB   |
+| passage-ranker.pistachio        | de, en, es, fr, it, ja, nl, pt, zs, pl | 0.380     | 358 ms         | 1070 MiB   |
+
+
+## Answer Finder
+
+Answer Finder are extractive question answering models developed by Sinequa. Given a query and a passage, they produce two lists of logit scores corresponding 
+to the start token and end token of an answer.
+
+Here is an overview of the model we deliver publicly here.
+
+| Model                          | Languages       | de   | en   | es   | fr   | ja   |
+|--------------------------------|-----------------|------|------|------|------|------|
+| answer-finder-v1-S-en          | en              | 70.6 | 79.5 | 54.1 | 0.5  | X    |
+| answer-finder-v1-L-multilingual | de, en, es, fr | 90.8 | 75.0 | 67.1 | 73.4 | X    |
+| answer-finder.yuzu             | ja              | X    | X    | X    | X    | 91.5 |
+
+| Model                          | Inference Time | GPU Memory |
+|--------------------------------|----------------|------------|
+| answer-finder-v1-S-en          | 128 ms         | 560 MiB    |
+| answer-finder-v1-L-multilingual | 362 ms         | 1060 MiB   |
+| answer-finder.yuzu             | 361 ms         | 1320 MiB   |