Remove mlflow dependency (#23)

Justfile

@@ -20,14 +20,14 @@ install:
     just python -m pip install types-requests
 
 format:
-    . ./activate ${VENV_NAME} && autoflake ${PROJECT_DIR} --remove-all-unused-imports --quiet --in-place -r --exclude third_party --exclude ultravox/model/gazelle
+    . ./activate ${VENV_NAME} && autoflake ${PROJECT_DIR} --remove-all-unused-imports --quiet --in-place -r --exclude third_party
     . ./activate ${VENV_NAME} && isort ${PROJECT_DIR} --force-single-line-imports
     . ./activate ${VENV_NAME} && black ${PROJECT_DIR}
 
 check:
     . ./activate ${VENV_NAME} && black ${PROJECT_DIR} --check
     . ./activate ${VENV_NAME} && isort ${PROJECT_DIR} --check --force-single-line-imports
-    . ./activate ${VENV_NAME} && autoflake  ${PROJECT_DIR} --check --quiet --remove-all-unused-imports -r --exclude third_party --exclude ultravox/model/gazelle
+    . ./activate ${VENV_NAME} && autoflake  ${PROJECT_DIR} --check --quiet --remove-all-unused-imports -r --exclude third_party
     . ./activate ${VENV_NAME} && mypy ${PROJECT_DIR}    
 
 test *ARGS=".":

mcloud.yaml

@@ -1,6 +1,6 @@
-# Gazelle POC training configuration
+# Ultravox POC training configuration
 
-name: gazelle-poc
+name: ultravox
 image: mosaicml/composer:latest
 compute:
   gpus: 8

mypy.ini

@@ -1,5 +1,3 @@
 [mypy]
 ignore_missing_imports = True
 
-[mypy-ultravox/model/gazelle.*]
-ignore_errors = True

requirements.txt

@@ -22,6 +22,3 @@ jiwer
 # Monitoring
 tensorboardx
 wandb
-neptune
-mlflow
-

ultravox/model/ultravox_processing.py

@@ -212,7 +212,7 @@ def _process(self, sample: datasets.VoiceSample) -> Dict[str, Any]:
             sample.messages, tokenize=False
         )
 
-        # Process audio and text using GazelleProcessor.
+        # Process audio and text using UltravoxProcessor.
         # Audio is expanded to be a [C x M] array, although C=1 for mono audio.
         audio = (
             np.expand_dims(sample.audio, axis=0) if sample.audio is not None else None

ultravox/training/configs/llama3_whisper.yaml

@@ -1,4 +1,4 @@
-# SLM with gazelle & llama3
+# SLM with ultravox & llama3
 exp_name: "llama3_whisper_s"
 
 # Make sure to accept the license agreement on huggingface hub

ultravox/training/train.py

@@ -8,7 +8,6 @@
 from typing import List, Optional
 
 import datasets as hf_datasets
-import mlflow
 import safetensors.torch
 import simple_parsing
 import torch
@@ -20,7 +19,6 @@
 
 from ultravox.data import datasets
 from ultravox.inference import infer
-from ultravox.inference import ultravox_infer
 from ultravox.model import ultravox_config
 from ultravox.model import ultravox_model
 from ultravox.model import ultravox_processing
@@ -33,17 +31,6 @@
 OUTPUT_EXAMPLE = {"text": "Hello, world!"}
 
 
-class GazelleMlflowWrapper(mlflow.pyfunc.PythonModel):
-    def predict(self, context, model_input):
-        sample = datasets.VoiceSample.from_prompt_and_buf(
-            model_input["text"], model_input["audio"]
-        )
-        return self.inference.infer(sample)
-
-    def load_context(self, context):
-        self.inference = ultravox_infer.UltravoxInference(context.artifacts["model_id"])
-
-
 def fix_hyphens(arg: str):
     return re.sub(r"^--([^=]+)", lambda m: "--" + m.group(1).replace("-", "_"), arg)
 
@@ -143,12 +130,6 @@ def main() -> None:
             dir="runs",
         )
 
-    # Starting MLflow; we need to set the experiment name before training starts.
-    if "mlflow" in args.report_logs_to and is_master:
-        mlflow.set_tracking_uri("runs/mlruns")
-        db_exp_name = f"/Shared/{args.exp_name}"
-        mlflow.set_experiment(db_exp_name)
-
     if args.model_load_dir:
         logging.info(f"Loading model state dict from {args.model_load_dir}")
         load_path = args.model_load_dir
@@ -274,20 +255,6 @@ def main() -> None:
     )
     trainer.train()
     trainer.save_model(args.output_dir)
-    if "mlflow" in args.report_logs_to and is_master:
-        signature = mlflow.models.signature.infer_signature(
-            INPUT_EXAMPLE, OUTPUT_EXAMPLE
-        )
-        model_info = mlflow.pyfunc.log_model(
-            python_model=GazelleMlflowWrapper(),
-            artifact_path="model",
-            pip_requirements="requirements.txt",
-            registered_model_name="ultravox",
-            input_example=INPUT_EXAMPLE,
-            signature=signature,
-        )
-        logging.info(f"Model logged to MLflow: {model_info.model_uri}")
-
     t_end = datetime.now()
     logging.info(f"end time: {t_end}")
     logging.info(f"elapsed: {t_end - t_start}")