Use the CLI in the tutorials (Lightning-AI#1094)

README.md

@@ -102,7 +102,7 @@ To generate text predictions, you need to download the model weights. **If you d
 Run inference:
 
 ```bash
-python litgpt/generate/base.py --prompt "Hello, my name is"
+litgpt generate base --prompt "Hello, my name is"
 ```
 
 This will run the 3B pretrained model and require ~7 GB of GPU memory using the `bfloat16` datatype.
@@ -112,7 +112,7 @@ This will run the 3B pretrained model and require ~7 GB of GPU memory using the
 You can also chat with the model interactively:
 
 ```bash
-python litgpt/chat/base.py
+litgpt chat
 ```
 
  
@@ -131,19 +131,19 @@ For example, you can either use
 Adapter ([Zhang et al. 2023](https://arxiv.org/abs/2303.16199)):
 
 ```bash
-python litgpt/finetune/adapter.py
+litgpt finetune adapter
 ```
 
 or Adapter v2 ([Gao et al. 2023](https://arxiv.org/abs/2304.15010)):
 
 ```bash
-python litgpt/finetune/adapter_v2.py
+litgpt finetune adapter_v2
 ```
 
 or LoRA ([Hu et al. 2021](https://arxiv.org/abs/2106.09685)):
 
 ```bash
-python litgpt/finetune/lora.py
+litgpt finetune lora
 ```
 
 (Please see the [tutorials/finetune_adapter](tutorials/finetune_adapter.md) for details on the differences between the two adapter methods.)

litgpt/utils.py

@@ -74,7 +74,7 @@ def check_valid_checkpoint_dir(checkpoint_dir: Path, lora: bool = False) -> None
     error_message = (
         f"--checkpoint_dir {str(checkpoint_dir.absolute())!r}{problem}."
         "\nFind download instructions at https://github.com/Lightning-AI/litgpt/blob/main/tutorials\n"
-        f"{extra}\nSee all download options by running:\n python litgpt/scripts/download.py"
+        f"{extra}\nSee all download options by running:\n litgpt download"
     )
     print(error_message, file=sys.stderr)
     raise SystemExit(1)

tests/test_utils.py

@@ -47,7 +47,7 @@ def test_check_valid_checkpoint_dir(tmp_path):
 Find download instructions at https://github.com/Lightning-AI/litgpt/blob/main/tutorials
 
 See all download options by running:
- python litgpt/scripts/download.py
+ litgpt download
     """.strip()
     assert out == expected
 
@@ -61,7 +61,7 @@ def test_check_valid_checkpoint_dir(tmp_path):
 Find download instructions at https://github.com/Lightning-AI/litgpt/blob/main/tutorials
 
 See all download options by running:
- python litgpt/scripts/download.py
+ litgpt download
     """.strip()
     assert out == expected
 
@@ -79,7 +79,7 @@ def test_check_valid_checkpoint_dir(tmp_path):
  --checkpoint_dir '{str(checkpoint_dir.absolute())}'
 
 See all download options by running:
- python litgpt/scripts/download.py
+ litgpt download
     """.strip()
     assert out == expected
 

tutorials/convert_hf_checkpoint.md

@@ -3,7 +3,7 @@
 By default, the `litgpt/scripts/download.py` script converts the downloaded HF checkpoint files into a LitGPT compatible format after downloading. For example,
 
 ```bash
-python litgpt/scripts/download.py --repo_id EleutherAI/pythia-14m
+litgpt download --repo_id EleutherAI/pythia-14m
 ```
 
 creates the following files:
@@ -28,7 +28,7 @@ To disable the automatic conversion, which is useful for development and debuggi
 ```bash
 rm -rf checkpoints/EleutherAI/pythia-14m 
 
-python litgpt/scripts/download.py \
+litgpt download \
   --repo_id EleutherAI/pythia-14m \
   --convert_checkpoint false
 
@@ -49,7 +49,7 @@ ls checkpoints/EleutherAI/pythia-14m
 The required files `lit_config.json` and `lit_model.pth` files can then be manually generated via the `litgpt/scripts/convert_hf_checkpoint.py` script:
 
 ```bash
-python litgpt/scripts/convert_hf_checkpoint.py \
+litgpt convert to_litgpt \
   --checkpoint_dir checkpoints/EleutherAI/pythia-14m
 ```
 

tutorials/convert_lit_models.md

@@ -5,7 +5,7 @@ LitGPT weights need to be converted to a format that Hugging Face understands wi
 We provide a helpful script to convert models LitGPT models back to their equivalent Hugging Face Transformers format:
 
 ```sh
-python litgpt/scripts/convert_lit_checkpoint.py \
+litgpt convert from_litgpt \
     --checkpoint_dir checkpoint_dir \
     --output_dir converted_dir
 ```
@@ -47,7 +47,7 @@ model = AutoModel.from_pretrained("online_repo_id", state_dict=state_dict)
 Please note that if you want to convert a model that has been fine-tuned using an adapter like LoRA, these weights should be [merged](../litgpt/scripts/merge_lora.py) to the checkpoint prior to converting.
 
 ```sh
-python litgpt/scripts/merge_lora.py \
+litgpt merge_lora \
     --checkpoint_dir path/to/lora/checkpoint_dir
 ```
 
@@ -73,7 +73,7 @@ by running `litgpt/scripts/download.py` without any additional arguments.
 Then, we download the model we specified via `$repo_id` above:
 
 ```bash
-python litgpt/scripts/download.py --repo_id $repo_id
+litgpt download --repo_id $repo_id
 ```
 
 2. Finetune the model:
@@ -82,7 +82,7 @@ python litgpt/scripts/download.py --repo_id $repo_id
 ```bash
 export finetuned_dir=out/lit-finetuned-model
 
-python litgpt/finetune/lora.py \
+litgpt finetune lora \
    --checkpoint_dir checkpoints/$repo_id \
    --out_dir $finetuned_dir \
    --train.epochs 1 \
@@ -94,15 +94,15 @@ python litgpt/finetune/lora.py \
 Note that this step only applies if the model was finetuned with `lora.py` above and not when `full.py` was used for finetuning.
 
 ```bash
-python litgpt/scripts/merge_lora.py \
+litgpt merge_lora \
     --checkpoint_dir $finetuned_dir/final
 ```
 
 
 4. Convert the finetuning model back into a HF format:
 
 ```bash
-python litgpt/scripts/convert_lit_checkpoint.py \
+litgpt convert from_litgpt \
    --checkpoint_dir $finetuned_dir/final/ \
    --output_dir out/hf-tinyllama/converted \
 ```

tutorials/download_model_weights.md

@@ -11,7 +11,7 @@ LitGPT supports a variety of LLM architectures with publicly available weights.
 To see all supported models, run the following command without arguments:
 
 ```bash
-python litgpt/scripts/download.py
+litgpt download
 ```
 
 The output is shown below:
@@ -128,7 +128,7 @@ Trelis/Llama-2-7b-chat-hf-function-calling-v2
 To download the weights for a specific model, use the `--repo_id` argument. Replace `<repo_id>` with the model's repository ID. For example:
 
 ```bash
-python litgpt/scripts/download.py --repo_id <repo_id>
+litgpt download --repo_id <repo_id>
 ```
 This command downloads the model checkpoint into the `checkpoints/` directory.
 
@@ -139,7 +139,7 @@ This command downloads the model checkpoint into the `checkpoints/` directory.
 For more options, add the `--help` flag when running the script:
 
 ```bash
-python litgpt/scripts/download.py --help
+litgpt download --help
 ```
 
 &nbsp;
@@ -148,7 +148,7 @@ python litgpt/scripts/download.py --help
 After conversion, run the model with the `--checkpoint_dir` flag, adjusting `repo_id` accordingly:
 
 ```bash
-python litgpt/chat/base.py --checkpoint_dir checkpoints/<repo_id>
+litgpt chat --checkpoint_dir checkpoints/<repo_id>
 ```
 
 &nbsp;
@@ -159,7 +159,7 @@ This section shows a typical end-to-end example for downloading and using TinyLl
 1. List available TinyLlama checkpoints:
 
 ```bash
-python litgpt/scripts/download.py | grep Tiny
+litgpt download | grep Tiny
 ```
 
 ```
@@ -171,13 +171,13 @@ TinyLlama/TinyLlama-1.1B-Chat-v1.0
 
 ```bash
 export repo_id=TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
-python litgpt/scripts/download.py --repo_id $repo_id
+litgpt download --repo_id $repo_id
 ```
 
 3. Use the TinyLlama model:
 
 ```bash
-python litgpt/chat/base.py --checkpoint_dir checkpoints/$repo_id
+litgpt chat --checkpoint_dir checkpoints/$repo_id
 ```
 
 &nbsp;
@@ -190,7 +190,7 @@ For example, to get access to the Gemma 2B model, you can do so by following the
 Once you've been granted access and obtained the access token you need to pass the additional `--access_token`:
 
 ```bash
-python litgpt/scripts/download.py \
+litgpt download \
   --repo_id google/gemma-2b \
   --access_token your_hf_token
 ```
@@ -203,7 +203,7 @@ The `download.py` script will automatically convert the downloaded model checkpo
 
 
 ```bash
-python litgpt/scripts/download.py \
+litgpt download \
   --repo_id <repo_id>
   --dtype bf16-true
 ```
@@ -218,15 +218,15 @@ For development purposes, for example, when adding or experimenting with new mod
 You can do this by passing the `--convert_checkpoint false` option to the download script:
 
 ```bash
-python litgpt/scripts/download.py \
+litgpt download \
   --repo_id <repo_id> \
   --convert_checkpoint false
 ```
 
 and then calling the `convert_hf_checkpoint.py` script:
 
 ```bash
-python litgpt/scripts/convert_hf_checkpoint.py \
+litgpt convert to_litgpt \
   --checkpoint_dir checkpoint_dir/<repo_id>
 ```
 
@@ -236,15 +236,15 @@ python litgpt/scripts/convert_hf_checkpoint.py \
 In some cases we don't need the model weight, for example, when we are pretraining a model from scratch instead of finetuning it. For cases like this, you can use the `--tokenizer_only` flag to only download a model's tokenizer, which can then be used in the pretraining scripts:
 
 ```bash
-python litgpt/scripts/download.py \
+litgpt download \
   --repo_id TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T \
   --tokenizer_only true
 ```
 
 and
 
 ```bash
-python litgpt/pretrain.py \
+litgpt pretrain \
   --data ... \
   --model_name tiny-llama-1.1b \
   --tokenizer_dir checkpoints/TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T/

tutorials/finetune_adapter.md

@@ -22,15 +22,15 @@ For more information about dataset preparation, also see the [prepare_dataset.md
 ## Running the finetuning
 
 ```bash
-python litgpt/finetune/adapter.py \
+litgpt finetune adapter \
   --data Alpaca \
   --checkpoint_dir checkpoints/stabilityai/stablelm-base-alpha-3b
 ```
 
 or for Adapter V2
 
 ```bash
-python litgpt/finetune/adapter_v2.py \
+litgpt finetune adapter_v2 \
   --data Alpaca \
   --checkpoint_dir checkpoints/stabilityai/stablelm-base-alpha-3b
 ```
@@ -49,15 +49,15 @@ For example, the following settings will let you finetune the model in under 1 h
 This script will save checkpoints periodically to the `out_dir` directory. If you are finetuning different models or on your own dataset, you can specify an output directory with your preferred name:
 
 ```bash
-python litgpt/finetune/adapter.py \
+litgpt finetune adapter \
   --data Alpaca \
   --out_dir out/adapter/my-model-finetuned
 ```
 
 or for Adapter V2
 
 ```bash
-python litgpt/finetune/adapter_v2.py \
+litgpt finetune adapter_v2 \
   --data Alpaca \
   --out_dir out/adapter_v2/my-model-finetuned
 ```
@@ -66,7 +66,7 @@ If your GPU does not support `bfloat16`, you can pass the `--precision 32-true`
 For instance, to fine-tune on MPS (the GPU on modern Macs), you can run
 
 ```bash
-python litgpt/finetune/adapter.py \
+litgpt finetune adapter \
   --data Alpaca \
   --out_dir out/adapter/my-model-finetuned \
   --precision 32-true
@@ -79,13 +79,13 @@ Note that `mps` as the accelerator will be picked up automatically by Fabric whe
 Optionally, finetuning using quantization can be enabled via the `--quantize` flag, for example using the 4-bit NormalFloat data type:
 
 ```bash
-python litgpt/finetune/adapter.py --quantize "bnb.nf4"
+litgpt finetune adapter --quantize "bnb.nf4"
 ```
 
 or using adapter_v2 with double-quantization:
 
 ```bash
-python litgpt/finetune/adapter_v2.py --quantize "bnb.nf4-dq"
+litgpt finetune adapter_v2 --quantize "bnb.nf4-dq"
 ```
 
 For additional benchmarks and resource requirements, please see the [Resource Tables](resource-tables.md).
@@ -95,15 +95,15 @@ For additional benchmarks and resource requirements, please see the [Resource Ta
 You can test the finetuned model with your own instructions by running:
 
 ```bash
-python litgpt/generate/adapter.py \
+litgpt generate adapter \
     --prompt "Recommend a movie to watch on the weekend." \
     --checkpoint_dir checkpoints/stabilityai/stablelm-base-alpha-3b
 ```
 
 or for Adapter V2
 
 ```bash
-python litgpt/generate/adapter_v2.py \
+litgpt generate adapter_v2 \
     --prompt "Recommend a movie to watch on the weekend." \
     --checkpoint_dir checkpoints/stabilityai/stablelm-base-alpha-3b
 ```
@@ -138,7 +138,7 @@ You can easily train on your own instruction dataset saved in JSON format.
 2. Run `litgpt/finetune/adapter.py` or `litgpt/finetune/adapter_v2.py` by passing in the location of your data (and optionally other parameters):
 
     ```bash
-    python litgpt/finetune/adapter.py \
+    litgpt finetune adapter \
         --data JSON \
         --data.json_path data/mydata.json \
         --checkpoint_dir checkpoints/tiiuae/falcon-7b \