update

README.md

@@ -12,7 +12,9 @@ docker run -it --gpus '"device=1"' --ipc=host --net=host -v $PWD:/app fine-tune-
 
 modal token set --token-id ak-r6mjZ61XGQtNoCDZGHrFLP --token-secret as-2m1UyDMKKwTo2uApJVGovn 
 
+https://modal.com/docs/reference/modal.config
+
 export MODAL_TOKEN_ID=ak-r6mjZ61XGQtNoCDZGHrFLP
 export MODAL_TOKEN_SECRET=as-2m1UyDMKKwTo2uApJVGovn 
 
-modal run
+modal run text2sql_training/modal_training.py

text2sql_training/llm_stf.py

@@ -23,19 +23,16 @@
 from datasets import load_dataset
 from random import randint
 from huggingface_hub import hf_hub_download
-
+from collections import defaultdict
 
 HF_TOKEN_READ = os.getenv("HF_TOKEN_READ")
 HF_TOKEN_WRITE = os.getenv("HF_TOKEN_WRITE")
 
 
 class DataConfig(Config):
-    # dataset_name: str = "motherduckdb/duckdb-text2sql-25k"
     dataset_name: str = "b-mc2/sql-create-context"
-
     train_data_path: str = "train_dataset-sql.json"
     test_data_path: str = "test_dataset-sql.json"
-
     test_size: float = 0.1
     sample_training: int = 5000
 
@@ -113,7 +110,7 @@ def test_data(context: AssetExecutionContext, create_text_to_sql_dataset):
 
     return dataset
 
-def run_training(pretrained_model_id: str, peft_model_id: str, train_data):
+def run_training(pretrained_model_id: str, peft_model_id: str, train_data) -> str:
     bnb_config = BitsAndBytesConfig(
         load_in_4bit=True,
         bnb_4bit_use_double_quant=True,
@@ -215,36 +212,29 @@ def run_training(pretrained_model_id: str, peft_model_id: str, train_data):
 
     return hub_model_id
 
-@asset(group_name="model", compute_kind="modal")
-def trained_model(
-    context: AssetExecutionContext, config: ModelTrainingConfig, train_data
-):
-
+@asset(group_name="model", compute_kind="modal-labs")
+def trained_model(context: AssetExecutionContext, config: ModelTrainingConfig, train_data):
     hub_model_id = run_training(pretrained_model_id=config.pretrained_model_id, peft_model_id=config.peft_model_id, train_data=train_data)
-
-    context.add_output_metadata(
-        {
-            "model_url": MetadataValue.url(f"https://huggingface.co/{hub_model_id}"),
-        }
-    )
+    context.add_output_metadata({"model_url": MetadataValue.url(f"https://huggingface.co/{hub_model_id}")})
+    return hub_model_id
 
 
 
 
 
-@asset(group_name="model")
+@asset(group_name="model", compute_kind="python")
 def model_card(context: AssetExecutionContext, trained_model):
+
     model_card_path = hf_hub_download(repo_id=trained_model, filename="README.md")
+
     with open(model_card_path, "r") as f:
         content = f.read()
-    context.add_output_metadata(
-        {
-            "content": MetadataValue.md(content),    
-        })
+
+    context.add_output_metadata({"content": MetadataValue.md(content)})
     return content
 
 
-@asset(group_name="model", compute_kind="modal")
+@asset(group_name="model", compute_kind="python")
 def test_results(context: AssetExecutionContext, test_data, trained_model, config: ModelTrainingConfig):
     tokenizer = AutoTokenizer.from_pretrained(config.peft_model_id)
     model = AutoPeftModelForCausalLM.from_pretrained(
@@ -262,9 +252,9 @@ def test_results(context: AssetExecutionContext, test_data, trained_model, confi
         pipe.tokenizer.convert_tokens_to_ids("<|eot_id|>")
     ]
 
-
+    results = defaultdict(list)
     number_of_eval_samples = 100
-    for s in test_data.shuffle().select(range(number_of_eval_samples)):
+    for s in test_data.select(range(number_of_eval_samples)):
         query = s['messages'][1]['content']
 
 
@@ -277,33 +267,16 @@ def test_results(context: AssetExecutionContext, test_data, trained_model, confi
         original_sql = s['messages'][2]['content'].lower()
         generated_sql = outputs[0]['generated_text'][len(prompt):].strip().lower()
 
-        inference_samples.append({
-            'query': test_data[rand_idx]['messages'][1]['content'],
-            'original_sql': original_sql,
-            'generated_sql': generated_sql,
-            'match': original_sql == generated_sql
-        })
+        results['query'].append(query)
+        results['original_sql'].append(original_sql)
+        results['generated_sql'].append(generated_sql)
+        results['hard_match'].append(original_sql == generated_sql)
 
 
     rouge = evaluate.load('rouge')
-    rouge.compute(predictions=original_sql, references=generated_sql)
-
-    # def evaluate_sample(sample):
-    #     prompt = pipe.tokenizer.apply_chat_template(sample["messages"][:2], tokenize=False, add_generation_prompt=True)
-    #     outputs = pipe(prompt, max_new_tokens=256, do_sample=False, temperature=0.1, top_k=50, top_p=0.1, eos_token_id=terminators, pad_token_id=pipe.tokenizer.pad_token_id)
-    #     generated_sql = outputs[0]['generated_text'][len(prompt):].strip().lower()
-    #     original_sql = sample["messages"][2]["content"].lower()
-    #     return 1 if generated_sql == original_sql else 0
-
-    success_rate = []
-    number_of_eval_samples = 100
-    # iterate over eval dataset and predict
-    for s in test_data.shuffle().select(range(number_of_eval_samples)):
-        success_rate.append(evaluate_sample(s))
-
-    # compute accuracy
-    accuracy = sum(success_rate)/len(success_rate)
+    rouge.compute(predictions=results['generated_sql'], references=results['original_sql'])
 
+
     context.add_output_metadata(
         {
             "inference_samples": MetadataValue.json(inference_samples),

text2sql_training/modal_training.py

@@ -1,22 +1,27 @@
 import modal
 from modal import Image
+from datasets import load_dataset
+from datasets import disable_caching
+import pandas as pd
+import os 
 
 app = modal.App("example-hello-world")
-custom_image = Image.from_registry("ghcr.io/kyryl-opens-ml/fine-tune-llm-in-2024:main")
+custom_image = Image.from_registry("ghcr.io/kyryl-opens-ml/fine-tune-llm-in-2024:main").env({"HF_TOKEN": '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', "HF_TOKEN_WRITE": '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'})
 
+# ENV HF_TOKEN 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
+# ENV HF_TOKEN_WRITE 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
 
-@app.function(image=custom_image, gpu="A100")
-def foo():
-    from datasets import load_dataset
-    import text2sql_training
+@app.function(image=custom_image, gpu="A100", mounts=[modal.Mount.from_local_python_packages("text2sql_training", "text2sql_training")], timeout=15 * 60)
+def run_training_modal(train_data_pandas: pd.DataFrame):
+    from datasets import Dataset
     from text2sql_training.llm_stf import run_training
 
-    train_data = load_dataset("json", data_files='train_dataset-sql.json', split="train")
-    model_url = run_training(pretrained_model_id='meta-llama/Meta-Llama-3-8B-Instruct', peft_model_id='modal-test', train_data=train_data)
+    model_url = run_training(pretrained_model_id='meta-llama/Meta-Llama-3-8B-Instruct', peft_model_id='modal-test', train_data=Dataset.from_pandas(train_data_pandas))
     return model_url
 
 
 @app.local_entrypoint()
 def main():
-    result = foo.remote()
+    train_data = load_dataset("json", data_files='train_dataset-sql.json', split="train")
+    result = run_training_modal.remote(train_data_pandas=train_data.to_pandas())
     print(result)