Skip to content

Convert FT scripts from python to jaclang #10

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Convert FT scripts from python to jaclang #10

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
263 changes: 263 additions & 0 deletions src/jac-backend/FT_script.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,263 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "8321d173",
"metadata": {
"vscode": {
"languageId": "plaintext"
}
},
"outputs": [],
"source": [
"!pip install virtualenv"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0a5cd32a",
"metadata": {
"vscode": {
"languageId": "plaintext"
}
},
"outputs": [],
"source": [
"!virtualenv myenv"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2508b93d",
"metadata": {},
"outputs": [],
"source": [
"!myenv/bin/pip install ipykernel"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b887351a",
"metadata": {},
"outputs": [],
"source": [
"!myenv/bin/python -m ipykernel install --user --name=myenv --display-name \"Python (myenv)\""
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ef7f058c",
"metadata": {},
"outputs": [],
"source": [
"!python -m pip install -U jaclang\n",
"!pip install torch\n",
"!pip install datasets\n",
"!pip install transformers\n",
"!pip install trl\n",
"!pip install bert_score\n",
"!pip install unsloth\n",
"!pip install -U typing_extensions\n",
"!pip install --upgrade typing_extensions\n",
"!pip install tomli"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e3027836",
"metadata": {},
"outputs": [],
"source": [
"%%writefile main.jac\n",
"import:py from unsloth, FastVisionModel ;\n",
"import:py torch ;\n",
"import:py from unsloth, is_bf16_supported ;\n",
"import:py from unsloth.trainer, UnslothVisionDataCollator ;\n",
"import:py from trl, SFTTrainer, SFTConfig ;\n",
"import:py from datasets, load_dataset ;\n",
"import:py from bert_score, score ;\n",
"import:py numpy as np ;\n",
"import:py from transformers, TextStreamer ;\n",
"\n",
"can load_model() {\n",
"\n",
" (model, tokenizer) = FastVisionModel.from_pretrained(\n",
" 'unsloth/Qwen2-VL-7B-Instruct',\n",
" load_in_4bit=True,\n",
" use_gradient_checkpointing='unsloth'\n",
" ); \n",
" model = FastVisionModel.get_peft_model(\n",
" model,\n",
" finetune_vision_layers=True,\n",
" finetune_language_layers=True,\n",
" finetune_attention_modules=True,\n",
" finetune_mlp_modules=True,\n",
" r=16,\n",
" lora_alpha=16,\n",
" lora_dropout=0,\n",
" bias='none',\n",
" random_state=3407,\n",
" use_rslora=False,\n",
" loftq_config=None\n",
" ); \n",
" return (model, tokenizer) ;\n",
"}\n",
"\n",
"can prep_train_dataset(num_img: Any) {\n",
"\n",
" instruction = 'Write the LaTeX representation for this image.'; \n",
" can convert_to_conversation(sample: Any) {\n",
"\n",
" conversation = [\n",
" {\n",
" 'role': 'user',\n",
" 'content': [{'type': 'text', 'text': instruction}, {'type': 'image', 'image': sample['image']}]\n",
" },\n",
" {\n",
" 'role': 'assistant',\n",
" 'content': [{'type': 'text', 'text': sample['text']}]\n",
" }\n",
" ]; \n",
" return {'messages': conversation} ;\n",
" }\n",
" \n",
" dataset = load_dataset('unsloth/LaTeX_OCR', split='train'); \n",
" dataset = dataset.select(range(num_img)); \n",
" converted_dataset = [convert_to_conversation(sample) for sample in dataset]; \n",
" return converted_dataset ;\n",
"}\n",
"\n",
"can prep_train_model(model: Any, tokenizer: Any, converted_dataset: Any, num_epochs: Any=5) {\n",
"\n",
" FastVisionModel.for_training(model) ; \n",
" trainer = SFTTrainer(\n",
" model=model,\n",
" tokenizer=tokenizer,\n",
" data_collator=UnslothVisionDataCollator(model, tokenizer),\n",
" train_dataset=converted_dataset,\n",
" args=SFTConfig(\n",
" per_device_train_batch_size=2,\n",
" gradient_accumulation_steps=4,\n",
" warmup_steps=5,\n",
" #max_steps=30,\n",
" num_train_epochs=num_epochs,\n",
" learning_rate=0.0002,\n",
" fp16=not is_bf16_supported(),\n",
" bf16=is_bf16_supported(),\n",
" logging_steps=1,\n",
" optim='adamw_8bit',\n",
" weight_decay=0.01,\n",
" lr_scheduler_type='linear',\n",
" seed=3407,\n",
" output_dir='outputs',\n",
" report_to='none',\n",
" remove_unused_columns=False,\n",
" dataset_text_field='',\n",
" dataset_kwargs={'skip_prepare_dataset': True},\n",
" dataset_num_proc=4,\n",
" max_seq_length=2048\n",
" )\n",
" ); \n",
" return trainer ;\n",
"}\n",
"\n",
"can get_response(test_dataset: Any, model: Any, tokenizer: Any, n: Any) {\n",
"\n",
" FastVisionModel.for_inference(model) ; \n",
" image = test_dataset[n]['image']; \n",
" instruction = 'Write the LaTeX representation for this image.'; \n",
" messages = [\n",
" {\n",
" 'role': 'user',\n",
" 'content': [{'type': 'image'}, {'type': 'text', 'text': instruction}]\n",
" }\n",
" ]; \n",
" input_text = tokenizer.apply_chat_template(\n",
" messages,\n",
" add_generation_prompt=True\n",
" ); \n",
" inputs = tokenizer(\n",
" image,\n",
" input_text,\n",
" add_special_tokens=False,\n",
" return_tensors='pt'\n",
" ).to('cuda'); \n",
" text_streamer = TextStreamer(tokenizer, skip_prompt=True); \n",
" output_id = model.generate(\n",
" **inputs,\n",
" streamer=text_streamer,\n",
" max_new_tokens=128,\n",
" use_cache=True,\n",
" temperature=1.5,\n",
" min_p=0.1 # use `top_p` instead of `min_p`\n",
" );\n",
" response = tokenizer.decode(output_id[0], skip_special_tokens=True); \n",
" return response ;\n",
"}\n",
"\n",
"can evaluate(response: Any, reference: Any) {\n",
"\n",
" (P, R, F1) = score(\n",
" [response],\n",
" [reference],\n",
" model_type='bert-base-uncased',\n",
" lang='en'\n",
" ); \n",
" return {\n",
" 'precision': P.mean().item(),\n",
" 'recall': R.mean().item(),\n",
" 'f1': F1.mean().item()\n",
" } ;\n",
"}\n",
"\n",
"with entry {\n",
"\n",
" (model, tokenizer) = load_model(); \n",
" train_dataset = prep_train_dataset(100); \n",
" trainer = prep_train_model(model, tokenizer, train_dataset); \n",
" trainer.train() ; \n",
" test_dataset = load_dataset('unsloth/LaTeX_OCR', split='test'); \n",
" response_dict = {'precision': 0, 'recall': 0, 'f1': 0}; \n",
" \n",
" for img in range(20) {\n",
" response = get_response(test_dataset, model, tokenizer, img); \n",
" reference = test_dataset[img]['text']; \n",
" accuracy = evaluate(response, reference); \n",
" response_dict['precision'] += accuracy['precision']; \n",
" response_dict['recall'] += accuracy['recall']; \n",
" response_dict['f1'] += accuracy['f1'];\n",
" }\n",
" \n",
" precision = (response_dict['precision'] / 20);\n",
" recall = (response_dict['recall'] / 20);\n",
" f1 = (response_dict['f1'] / 20);\n",
" \n",
" print(f\"{'VQA Accuracy: Precision: '}{precision}{', Recall: '}{recall}{', F1: '}{f1}\");\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e3a7c4ec",
"metadata": {},
"outputs": [],
"source": [
"!jac run main.jac"
]
}
],
"metadata": {
"language_info": {
"name": "python"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
Loading