Update README.md

README.md

@@ -1,59 +1,142 @@
 ---
-license: other
-base_model: laurentiubp/CataLLaMA-v0.2.0
+license: llama3
+base_model: catallama/CataLlama-v0.1-Instruct-SFT
 tags:
-- trl
-- dpo
-- generated_from_trainer
+- llama
+- llama-3
+- Catalan
 model-index:
-- name: CataLLaMA-v0.2.6
+- name: catallama/CataLlama-v0.1-Instruct-DPO
   results: []
+datasets:
+- catallama/Catalan-DPO
+language:
+- ca
+- en
+pipeline_tag: text-generation
 ---
 
-<!-- This model card has been generated automatically according to the information the Trainer had access to. You
-should probably proofread and complete it, then remove this comment. -->
+**catallama/CataLlama-v0.1-Instruct-DPO** is a DPO fine-tune of [catallama/CataLlama-v0.1-Instruct-SFT](https://huggingface.co/catallama/CataLlama-v0.1-Instruct-SFT) on the [catallama/Catalan-DPO](https://huggingface.co/datasets/catallama/Catalan-DPO) dataset.
 
-# CataLLaMA-v0.2.6
+The model shows improved proficiency with the Catalan language.
 
-This model is a fine-tuned version of [laurentiubp/CataLLaMA-v0.2.0](https://huggingface.co/laurentiubp/CataLLaMA-v0.2.0) on an unknown dataset.
+**This is an instruction fine-tuned model, optimised with DPO, proficient on the following tasks in Catalan**
 
-## Model description
+- *Information extraction (suitable for RAG)*
+- *Named Entity Recognition (NER)*
+- *Translation from English to Catalan and Catalan to English*
+- *Summarization - both short form and long form*
+- *Chat*
+- *Sentiment analysis*
+- *Open question answering*
 
-More information needed
+**Model developers** [Laurentiu Petrea](https://www.linkedin.com/in/laurentiupetrea/) based on Llama-3 from Meta.
 
-## Intended uses & limitations
+**Model Architecture** CataLlama is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and direct preference optimisation (DPO) to align with human preferences for helpfulness and safety.
 
-More information needed
+**License** The model uses the llama-3 license available at: [https://llama.meta.com/llama3/license](https://llama.meta.com/llama3/license)
 
-## Training and evaluation data
+### Use with transformers
+
+See the snippet below for usage with Transformers:
+
+**The model follows the same prompt template as Llama-3 Instruct**
+
+```python
+import transformers
+import torch
+
+model_id = "catallama/CataLlama-v0.1-Base"
+
+pipeline = transformers.pipeline(
+    "text-generation",
+    model=model_id,
+    model_kwargs={"torch_dtype": torch.bfloat16},
+    device_map="auto",
+)
+
+messages = [
+    {"role": "user", "content": "Ei com estàs avui?"},
+]
+
+prompt = pipeline.tokenizer.apply_chat_template(
+    messages, 
+    tokenize=False, 
+    add_generation_prompt=True
+)
+
+outputs = pipeline(
+    prompt,
+    max_new_tokens=1024,
+    do_sample=True,
+    temperature=0.6,
+    top_p=0.9,
+)
+
+print(outputs[0]["generated_text"][len(prompt):])
+```
+
+**Examples**
+
+```
+Prompt: Pot Donald Trump sopar amb Juli Cèsar?
+
+Answer: No, Donald Trump no podria sopar amb Juli Cèsar, ja que Juli Cèsar va morir fa més de 2.000 anys i, per tant, no està viu per a poder sopar amb ell. A més, Juli Cèsar va ser un personatge històric i polític de l'antiga Roma, mentre que Donald Trump és el 45è president dels Estats Units, personatges que pertanyen a dues èpoques i cultures diferents. Per tant, la idea de que poguessin sopar junts és absurda i no té cap sentit històric ni lògic.
+```
+
+```
+Prompt: A quina velocitat poden volar els cocodrils?
+
+Answer: Els cocodrils no poden volar, ja que no tenen ales ni cap estructura que els permeti elevar-se per l'aire. Aquestes criatures aquàtiques depenen de la seva força i agilitat per moure's i caçar preses a terra ferma o al riu. No obstant això, poden assolir velocitats notables quan es tracta de fugir o perseguir una presa. El cocodril més ràpid, el cocodril negre africà, pot arribar a assolir velocitats de fins a 40 km/h en terra ferma, tot i que la seva velocitat a l'aigua és molt menor.
+```
+
+```
+*Prompt*: Explica pas a pas com resoldre l'equació següent: 2x + 10 = 0
+
+*Answer*:
+Per resoldre l'equació 2x + 10 = 0, segueix els passos següents:
+
+1. Aïllar la variable: per fer-ho, hem de restar 10 de tots dos costats de l'equació. Així, tenim:
+
+2x = -10
+
+2. Dividir tots dos costats per 2: això ens permet obtenir una equació en forma d'una sola incògnita. Així, tenim:
+
+x = -5
+
+3. Comprovar: sempre és recomanable comprovar la solució en l'equació original. Substituint x = -5 en 2x + 10 = 0, obtenim:
+
+2(-5) + 10 = -10 + 10 = 0
+
+Com que l'equació es compleix, la solució x = -5 és vàlida.
+```
 
-More information needed
 
 ## Training procedure
 
+The model was trained **with the same prompt template of Llama-3 Instruct**.
+
+The model was trained for two epochs on **4x A100 80GB GPUs using DeepSpeed ZeRO** State-3 without CPU offloading.
+
+
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
 - learning_rate: 5e-07
-- train_batch_size: 4
-- eval_batch_size: 4
-- seed: 42
 - distributed_type: multi-GPU
 - num_devices: 4
-- total_train_batch_size: 16
-- total_eval_batch_size: 16
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: linear
 - lr_scheduler_warmup_steps: 100
 - num_epochs: 1
 
-### Training results
 
+## Intended Use
+
+**Note:** This model is not intended to beat benchmarks, but to demonstrate techniques for augmenting LLMs on new languages and preserve rare languages as part of our world heritage.
 
+**Intended Use Cases** Llama 3 is intended for commercial and research use in English. Instruction tuned models are intended for assistant-like chat, whereas pretrained models can be adapted for a variety of natural language generation tasks.
 
-### Framework versions
+**Out-of-scope** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and Llama 3 Community License. Use in languages other than English**.
 
-- Transformers 4.38.1
-- Pytorch 2.1.0+cu118
-- Datasets 2.16.1
-- Tokenizers 0.15.2
+**Note: Developers may fine-tune Llama 3 models for languages beyond English provided they comply with the Llama 3 Community License and the Acceptable Use Policy.