Create README.md

README.md

@@ -0,0 +1,118 @@
+---
+# For reference on model card metadata, see the spec: https://github.com/huggingface/hub-docs/blob/main/modelcard.md?plain=1
+# Doc / guide: https://huggingface.co/docs/hub/model-cards
+{}
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+This model aims to provide a Question Answering model tuned with a short (128 tokens per row) Question Answering dataset
+The dataset enables fine tuning in local with small HW, such as 1 GPU with 16 Go RAM
+
+
+## Model Details
+
+The model has been dowloaded from ibm/mpt-7b-instruct2 (Apache 2.0 License.) and tuned with Supervised Fine-tuning Trainer and PEFT LoRa
+
+### Model Description
+
+
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+
+### Direct Use
+
+text = "Below is an instruction from Human. Write a response.\n    ### Instruction:\n   How to diagnose Parasites - Baylisascaris infection ?\n    ### Response:"
+inputs = tokenizer(text, return_tensors="pt").to('cuda')
+out = model.generate(**inputs, max_new_tokens=100)
+
+print(tokenizer.decode(out[0]))
+
+
+## Bias, Risks, and Limitations
+
+In order to reduce training duration, the model has been trained only with the first 5100 rows of the 15500 rows dataset
+
+### Recommendations
+
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. 
+Generation of plausible yet incorrect factual information, termed hallucination, is an unsolved issue in large language models.
+
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+## Training Details
+
+per_device_train_batch_size = 1
+gradient_accumulation_steps = 16
+epoch = 5
+
+Step 	Training Loss
+64 	    1.618400
+128 	1.084200
+192 	1.021800
+256 	1.014300
+320 	0.960500
+384 	0.905900
+448 	0.885200
+512 	0.847400
+576 	0.889400
+640 	0.861000
+704 	0.800400
+768 	0.768600
+832 	0.750300
+896 	0.780200
+960 	0.762700
+1024 	0.698600
+1088 	0.672600
+1152 	0.693100
+1216 	0.708900
+1280 	0.662700
+1344 	0.630400
+1408 	0.624600
+1472 	0.627200
+1536 	0.628000
+1600 	0.603300
+
+### Training Data
+
+Laurent1/MedQuad-MedicalQnADataset_128tokens_max
+
+#### Preprocessing [optional]
+
+Dataset already preprocessed (128 tokens max and truncated at a sentence end to keep meaning)
+
+
+#### Training Hyperparameters
+
+
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype=torch.float16,
+)
+
+model = AutoModelForCausalLM.from_pretrained(
+        "ibm/mpt-7b-instruct2",
+        device_map="auto",
+        torch_dtype=torch.float16, #torch.bfloat16,
+        trust_remote_code=True
+            )
+
+#### Speeds, Sizes, Times [optional]
+
+Training :
+
+6287.4s - GPU T4 x2