Add model card

README.md

@@ -1,62 +1,55 @@
 ---
-library_name: peft
-base_model: AnnaelleMyriam/SFT_M3_model
+language: en
+license: apache-2.0
 tags:
-- mcqa
+- text-generation
 - question-answering
+- mcqa
+- merged
 - sft
 - lora
-- qwen
-- unsloth
-- generated_from_trainer
-model-index:
-- name: MNLP_M3_mcqa_model_test
-  results: []
+base_model: AnnaelleMyriam/SFT_M3_model
 ---
 
-<!-- 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. -->
-
-# MNLP_M3_mcqa_model_test
-
-This model is a fine-tuned version of [AnnaelleMyriam/SFT_M3_model](https://huggingface.co/AnnaelleMyriam/SFT_M3_model) on an unknown dataset.
-
-## Model description
-
-More information needed
+# MNLP M3 MCQA Merged Model
 
-## Intended uses & limitations
+This model is a merged version of:
+- **Base SFT Model**: `AnnaelleMyriam/SFT_M3_model`
+- **LoRA Adapter**: `aymanbakiri/MNLP_M3_mcqa_model_adapter`
 
-More information needed
+## Model Description
 
-## Training and evaluation data
+This is a specialized model for Multiple Choice Question Answering (MCQA) tasks, created by:
+1. Starting with the SFT model `AnnaelleMyriam/SFT_M3_model`
+2. Fine-tuning with LoRA adapters on MCQA data
+3. Merging the LoRA weights back into the base model
 
-More information needed
+## Usage
 
-## Training procedure
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
-### Training hyperparameters
+model = AutoModelForCausalLM.from_pretrained("aymanbakiri/MNLP_M3_mcqa_model_test")
+tokenizer = AutoTokenizer.from_pretrained("aymanbakiri/MNLP_M3_mcqa_model_test")
 
-The following hyperparameters were used during training:
-- learning_rate: 2e-05
-- train_batch_size: 4
-- eval_batch_size: 8
-- seed: 42
-- gradient_accumulation_steps: 2
-- total_train_batch_size: 8
-- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
-- lr_scheduler_type: cosine
-- lr_scheduler_warmup_ratio: 0.05
-- num_epochs: 1
+# Example usage for MCQA
+prompt = """Question: What is the capital of France?
+Options: (A) London (B) Berlin (C) Paris (D) Madrid
+Answer:"""
 
-### Training results
+inputs = tokenizer(prompt, return_tensors="pt")
+outputs = model.generate(**inputs, max_new_tokens=5)
+answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
+print(answer)
+```
 
+## Training Details
 
+- Base Model: SFT model fine-tuned for instruction following
+- LoRA Configuration: r=16, alpha=32, dropout=0.1
+- Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj, lm_head
+- Training Data: MNLP M2 MCQA Dataset
 
-### Framework versions
+## Performance
 
-- PEFT 0.15.2
-- Transformers 4.52.4
-- Pytorch 2.7.0+cu126
-- Datasets 3.6.0
-- Tokenizers 0.21.1
+This merged model should provide better performance than the original LoRA adapter while being easier to deploy and use.