Merge branch 'main' of github.com:Lightning-AI/litgpt into feature/lo…

…nglora

.github/workflows/publish.yaml

@@ -0,0 +1,39 @@
+# To create a release, create a tag and push it to GitHub:
+#git tag -a "v0.0.1-beta" -m "beta version testing"
+#git push --tags
+# https://dev.to/iamtekson/publish-package-to-pypi-and-release-new-version-using-github-actions-108k
+name: Publish LitGPT to PyPI
+
+on:
+  push:
+    tags:
+      - "v*"
+jobs:
+  build-n-publish:
+    name: Build and publish to PyPI
+    runs-on: ubuntu-latest
+    environment:
+      name: pypi
+      url: https://pypi.org/p/litgpt
+    permissions:
+      id-token: write
+
+    steps:
+      - name: Checkout source
+        uses: actions/checkout@v3
+
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: "3.x"
+
+      - name: Build source and wheel distributions
+        run: |
+          python -m pip install --upgrade build twine
+          python -m build
+          twine check --strict dist/*
+      - name: Publish distribution to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          user: __token__
+          password: ${{ secrets.PYPI_API_TOKEN }}

litgpt/chat/base.py

@@ -219,9 +219,3 @@ def main(
             file=sys.stderr,
         )
         fabric.print()
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(main)

litgpt/deploy/serve.py

@@ -162,7 +162,3 @@ def run_server(
         devices=devices)
 
     server.run(port=port)
-
-
-if __name__ == "__main__":
-    CLI(run_server)

litgpt/eval/evaluate.py

@@ -112,7 +112,3 @@ def convert_and_evaluate(
         torch_random_seed=seed,
     )
     prepare_results(results, save_filepath)
-
-
-if __name__ == "__main__":
-    CLI(convert_and_evaluate)

litgpt/finetune/adapter.py

@@ -22,7 +22,6 @@
 from litgpt.prompts import save_prompt_style
 from litgpt.tokenizer import Tokenizer
 from litgpt.utils import (
-    CLI,
     CycleIterator,
     check_valid_checkpoint_dir,
     choose_logger,
@@ -396,9 +395,3 @@ def validate_args(train: TrainArgs, eval: EvalArgs) -> None:
         issues.append(f"{__file__} requires either epochs or max_steps to be set. This is set in {train}")
     if issues:
         raise ValueError("\n".join(issues))
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(setup)

litgpt/finetune/adapter_v2.py

@@ -22,7 +22,6 @@
 from litgpt.prompts import save_prompt_style
 from litgpt.tokenizer import Tokenizer
 from litgpt.utils import (
-    CLI,
     CycleIterator,
     check_valid_checkpoint_dir,
     choose_logger,
@@ -396,9 +395,3 @@ def validate_args(train: TrainArgs, eval: EvalArgs) -> None:
         issues.append(f"{__file__} requires either epochs or max_steps to be set. This is set in {train}")
     if issues:
         raise ValueError("\n".join(issues))
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(setup)

litgpt/finetune/full.py

@@ -22,7 +22,6 @@
 from litgpt.prompts import save_prompt_style
 from litgpt.tokenizer import Tokenizer
 from litgpt.utils import (
-    CLI,
     CycleIterator,
     check_valid_checkpoint_dir,
     choose_logger,
@@ -420,9 +419,3 @@ def validate_longlora_args(config: Config, longlora: LongLoraArgs):
                 f"LongLora context length ({longlora.context_length}) must be a multiple of the number of groups "
                 f"({longlora.n_groups})."
             )
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(setup)

litgpt/finetune/lora.py

@@ -32,7 +32,6 @@
 from litgpt.scripts.merge_lora import merge_lora
 from litgpt.tokenizer import Tokenizer
 from litgpt.utils import (
-    CLI,
     CycleIterator,
     check_valid_checkpoint_dir,
     choose_logger,
@@ -502,10 +501,4 @@ def validate_longlora_args(config: Config, longlora: LongLoraArgs):
             raise ValueError(
                 f"LongLora context length ({longlora.context_length}) must be a multiple of the number of groups "
                 f"({longlora.n_groups})."
-            )
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(setup)
+            )

litgpt/generate/adapter.py

@@ -123,9 +123,3 @@ def main(
     fabric.print(f"\n\nTime for inference: {t:.02f} sec total, {tokens_generated / t:.02f} tokens/sec", file=sys.stderr)
     if fabric.device.type == "cuda":
         fabric.print(f"Memory used: {torch.cuda.max_memory_allocated() / 1e9:.02f} GB", file=sys.stderr)
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(main)

litgpt/generate/adapter_v2.py

@@ -13,7 +13,7 @@
 from litgpt.adapter_v2 import GPT, Config
 from litgpt.generate.base import generate
 from litgpt.prompts import has_prompt_style, load_prompt_style
-from litgpt.utils import CLI, check_valid_checkpoint_dir, get_default_supported_precision, lazy_load
+from litgpt.utils import check_valid_checkpoint_dir, get_default_supported_precision, lazy_load
 
 
 def main(
@@ -123,9 +123,3 @@ def main(
     fabric.print(f"\n\nTime for inference: {t:.02f} sec total, {tokens_generated / t:.02f} tokens/sec", file=sys.stderr)
     if fabric.device.type == "cuda":
         fabric.print(f"Memory used: {torch.cuda.max_memory_allocated() / 1e9:.02f} GB", file=sys.stderr)
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(main)

litgpt/generate/base.py

@@ -264,9 +264,3 @@ def main(
         )
     if fabric.device.type == "cuda":
         fabric.print(f"Memory used: {torch.cuda.max_memory_allocated() / 1e9:.02f} GB", file=sys.stderr)
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(main)

litgpt/generate/full.py

@@ -13,7 +13,7 @@
 from litgpt import GPT, Config, PromptStyle, Tokenizer
 from litgpt.generate.base import generate
 from litgpt.prompts import has_prompt_style, load_prompt_style
-from litgpt.utils import CLI, check_valid_checkpoint_dir, get_default_supported_precision, load_checkpoint
+from litgpt.utils import check_valid_checkpoint_dir, get_default_supported_precision, load_checkpoint
 
 
 def main(
@@ -141,9 +141,3 @@ def main(
     fabric.print(f"\n\nTime for inference: {t:.02f} sec total, {tokens_generated / t:.02f} tokens/sec", file=sys.stderr)
     if fabric.device.type == "cuda":
         fabric.print(f"Memory used: {torch.cuda.max_memory_allocated() / 1e9:.02f} GB", file=sys.stderr)
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(main)

litgpt/generate/sequentially.py

@@ -21,7 +21,7 @@
 import litgpt.generate.base as generate_base
 from litgpt import GPT, Config, Tokenizer
 from litgpt.model import Block, build_mask_cache
-from litgpt.utils import CLI, check_valid_checkpoint_dir, get_default_supported_precision
+from litgpt.utils import check_valid_checkpoint_dir, get_default_supported_precision
 
 
 @torch.inference_mode()
@@ -254,9 +254,3 @@ def main(
             f"Time for inference {i + 1}: {t:.02f} sec total, {tokens_generated / t:.02f} tokens/sec", file=sys.stderr
         )
     print(f"Memory used: {torch.cuda.max_memory_allocated() / 1e9:.02f} GB", file=sys.stderr)
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(main)

litgpt/generate/tp.py

@@ -248,9 +248,3 @@ def main(
         )
     if fabric.device.type == "cuda":
         fabric.print(f"Memory used: {torch.cuda.max_memory_allocated() / 1e9:.02f} GB", file=sys.stderr)
-
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(main)

litgpt/pretrain.py

@@ -443,8 +443,3 @@ def validate_args(train: TrainArgs, eval: EvalArgs, initial_checkpoint_dir, resu
     if issues:
         raise ValueError("\n".join(issues))
 
-
-if __name__ == "__main__":
-    torch.set_float32_matmul_precision("high")
-
-    CLI(setup)

litgpt/scripts/convert_hf_checkpoint.py

@@ -349,9 +349,3 @@ def convert_hf_checkpoint(
         gc.collect()
         print(f"Saving converted checkpoint to {checkpoint_dir}")
         saver.save(sd)
-
-
-if __name__ == "__main__":
-    from jsonargparse import CLI
-
-    CLI(convert_hf_checkpoint)

litgpt/scripts/convert_lit_checkpoint.py

@@ -10,7 +10,7 @@
 
 from litgpt import Config
 from litgpt.scripts.convert_hf_checkpoint import layer_template, load_param
-from litgpt.utils import CLI, incremental_save, lazy_load
+from litgpt.utils import incremental_save, lazy_load
 
 
 def copy_weights_falcon(
@@ -265,7 +265,3 @@ def convert_lit_checkpoint(checkpoint_dir: Path, output_dir: Path) -> None:
         copy_fn(sd, lit_weights, saver=saver)
         gc.collect()
         saver.save(sd)
-
-
-if __name__ == "__main__":
-    CLI(convert_lit_checkpoint)

litgpt/scripts/convert_pretrained_checkpoint.py

@@ -46,7 +46,3 @@ def convert_pretrained_checkpoint(checkpoint_dir: Path, output_dir: Path) -> Non
         saver.save(converted_state_dict)
 
     copy_config_files(checkpoint_dir, output_dir)
-
-
-if __name__ == "__main__":
-    CLI(convert_pretrained_checkpoint)

litgpt/scripts/download.py

@@ -144,7 +144,3 @@ def gated_repo_catcher(repo_id: str, access_token: Optional[str]):
                     f" visit https://huggingface.co/{repo_id} for more information."
                 ) from None
         raise e from None
-
-
-if __name__ == "__main__":
-    CLI(download_from_hub)

litgpt/scripts/merge_lora.py

@@ -86,7 +86,3 @@ def load_lora_metadata(checkpoint_dir: Path) -> Tuple[Dict[str, Any], Path, Opti
     pretrained_checkpoint_dir = Path(hparams["checkpoint_dir"])
     precision = hparams.get("precision")
     return lora_params, pretrained_checkpoint_dir, precision
-
-
-if __name__ == "__main__":
-    CLI(merge_lora)

litgpt/utils.py

@@ -444,6 +444,7 @@ def save_hyperparameters(function: callable, checkpoint_dir: Path) -> None:
         ("finetune", "lora"),
         ("finetune", "adapter"),
         ("finetune", "adapter_v2"),
+        ("finetune",),
         ("pretrain",),
     ]
     for known_command in known_commands:

pyproject.toml

@@ -37,7 +37,7 @@ all = [
     "sentencepiece>=0.2.0",      # llama-based models
     "tokenizers>=0.15.2",        # pythia, falcon, redpajama
     "requests>=2.31.0",          # litgpt.data
-    "litdata>=0.2.2",            # litgpt.data
+    "litdata>=0.2.2,<0.2.6",     # litgpt.data
     "litserve>=0.1.0",           # litgpt.deploy
     "zstandard>=0.22.0",         # litgpt.data.prepare_slimpajama.py
     "pandas>=1.9.0",             # litgpt.data.prepare_starcoder.py

tests/test_chat.py

@@ -122,13 +122,8 @@ def test_main(mocked_input, stop_iteration, fake_checkpoint_dir, monkeypatch, te
     assert re.match("Now chatting with Llama 3.*>> .*Reply: foo bar baz", out.getvalue(), re.DOTALL)
 
 
-@pytest.mark.parametrize("mode", ["file", "entrypoint"])
-def test_cli(mode):
-    if mode == "file":
-        cli_path = Path(__file__).parent.parent / "litgpt/chat/base.py"
-        args = [sys.executable, cli_path, "-h"]
-    else:
-        args = ["litgpt", "chat", "-h"]
+def test_cli():
+    args = ["litgpt", "chat", "-h"]
     output = subprocess.check_output(args)
     output = str(output.decode())
     assert "Starts a conversation" in output

tests/test_evaluate.py

@@ -1,14 +1,11 @@
 # Copyright Lightning AI. Licensed under the Apache License 2.0, see LICENSE file.
 
 import subprocess
-import sys
 from contextlib import redirect_stdout
 from dataclasses import asdict
 from io import StringIO
-from pathlib import Path
 from unittest import mock
 
-import pytest
 import torch
 import yaml
 
@@ -43,13 +40,8 @@ def test_evaluate_script(tmp_path):
     assert "Loading checkpoint shards" not in stdout
 
 
-@pytest.mark.parametrize("mode", ["file", "entrypoint"])
-def test_cli(mode):
-    if mode == "file":
-        cli_path = Path(__file__).parent.parent / "litgpt/eval/evaluate.py"
-        args = [sys.executable, cli_path, "-h"]
-    else:
-        args = ["litgpt", "evaluate", "-h"]
+def test_cli():
+    args = ["litgpt", "evaluate", "-h"]
     output = subprocess.check_output(args)
     output = str(output.decode())
     assert "run the LM Evaluation Harness" in output

tests/test_generate.py

@@ -83,13 +83,8 @@ def test_main(fake_checkpoint_dir, monkeypatch, tensor_like):
     assert "'padded_vocab_size': 512, 'n_layer': 2, 'n_head': 4" in err.getvalue()
 
 
-@pytest.mark.parametrize("mode", ["file", "entrypoint"])
-def test_cli(mode):
-    if mode == "file":
-        cli_path = Path(__file__).parent.parent / "litgpt/generate/base.py"
-        args = [sys.executable, cli_path, "-h"]
-    else:
-        args = ["litgpt", "generate", "base", "-h"]
+def test_cli():
+    args = ["litgpt", "generate", "base", "-h"]
     output = subprocess.check_output(args)
     output = str(output.decode())
     assert "Generates text samples" in output

tests/test_generate_adapter.py

@@ -48,13 +48,8 @@ def test_main(fake_checkpoint_dir, monkeypatch, version, tensor_like):
 
 
 @pytest.mark.parametrize("version", ("", "_v2"))
-@pytest.mark.parametrize("mode", ["file", "entrypoint"])
-def test_cli(version, mode):
-    if mode == "file":
-        cli_path = Path(__file__).parent.parent / f"litgpt/generate/adapter{version}.py"
-        args = [sys.executable, cli_path, "-h"]
-    else:
-        args = ["litgpt", "generate", f"adapter{version}", "-h"]
+def test_cli(version):
+    args = ["litgpt", "generate", f"adapter{version}", "-h"]
     output = subprocess.check_output(args)
     output = str(output.decode())
     assert "Generates a response" in output