Open source weights! (#97)

Summary:

Add code to download weights and demo code for running model.

Weights at:
- https://huggingface.co/collections/facebook/blt-6801263d4ac1704702a192a6
- https://huggingface.co/facebook/blt
- https://huggingface.co/facebook/blt-1b
- https://huggingface.co/facebook/blt-7b

Test Plan:

.gitignore

@@ -169,4 +169,4 @@ internal/
 jobs_parallel-copy/
 wandb/
 *.ipynb
-
+hf-weights/

README.md

@@ -46,7 +46,25 @@ Once that is done you can activate the environment
 conda activate blt_<date>
 ```
 
-use the provided script to download and prepare data from huggingface (among `fineweb_edu`, `fineweb_edu_10bt`, or `dclm_baseline_1.0`).
+## Downloading HF Model Weights and Generating Text
+
+We have released weights on HF for the [BLT 1B Model](https://huggingface.co/facebook/blt-1b) and [BLT 7B Model](https://huggingface.co/facebook/blt-7b).
+We are actively working with HF to make BLT available in [Transformers](https://huggingface.co/docs/transformers/en/index) and will update this when it is.
+In the meantime, you can follow these instructions to load model weights, initialize a model, and generate text.
+These instructions have been tested on H100 GPUs, but we can only offer suggestions on running on other hardware.
+
+1. On the model weights HF page, create a HuggingFace account, request access to weights, and wait for approval.
+2. On the huggingface cli, login: `huggingface-cli login`
+3. Download the model weights with: `python download_blt_weights.py`, which will load to `hf-weights`
+4. Run the generate demo: `python demo.py "A BLT has"`.
+
+The demo generates text, but is also a good starting point for loading BLT in your own code.
+
+## Downloading Training Data
+
+Note: The following instructions are not well tested in the BLT code as it is based on the lingua code, which we have diverged from.
+
+Use the provided script to download and prepare data from huggingface (among `fineweb_edu`, `fineweb_edu_10bt`, or `dclm_baseline_1.0`).
 This command will download the `fineweb_edu` and prepare it for training in the `./data` directory, specifying the amount of memory `terashuf` (the tool used to shuffle samples) will be allocated. By default, the number of chunks (`nchunks`) is 32. If you are running on fewer than 32 GPUs, it is recommended to set `nchunks` to 1 or to match `nchunks` with the number of GPUs (`nchunks` = NGPUs). See [here](https://github.com/facebookresearch/lingua/issues/55#issuecomment-2483643076) for more details.
 
 ```bash

bytelatent/data/patcher.py

@@ -476,12 +476,19 @@ class Patcher:
             assert (
                 patcher_args.entropy_model_checkpoint_dir is not None
             ), "Cannot require realtime patching without an entropy model checkpoint"
+            maybe_consolidated = os.path.join(
+                patcher_args.entropy_model_checkpoint_dir,
+                "consolidated/consolidated.pth",
+            )
+            if os.path.exists(maybe_consolidated):
+                state_path = maybe_consolidated
+            else:
+                state_path = os.path.join(
+                    patcher_args.entropy_model_checkpoint_dir, "consolidated.pth"
+                )
             entropy_model = load_entropy_model(
                 patcher_args.entropy_model_checkpoint_dir,
-                os.path.join(
-                    patcher_args.entropy_model_checkpoint_dir,
-                    "consolidated/consolidated.pth",
-                ),
+                state_path,
             )
             entropy_model, _ = to_device(entropy_model, patcher_args.patching_device)
             self.entropy_model = entropy_model

bytelatent/eval.py

@@ -206,7 +206,9 @@ def eval_ppl_on_path(
                 pred = model(x, patch_lengths=patch_lengths)
             else:
                 pred = model(x)
-            loss = F.cross_entropy(pred.flatten(0, 1), y.flatten(0, 1), reduction="sum", ignore_index=0)
+            loss = F.cross_entropy(
+                pred.flatten(0, 1), y.flatten(0, 1), reduction="sum", ignore_index=0
+            )
             total_loss += loss.item()
         else:
             raise NotImplementedError()

bytelatent/generate.py

@@ -25,7 +25,11 @@ from bytelatent.checkpoint import (
 )
 from bytelatent.config_parser import parse_args_to_pydantic_model
 from bytelatent.data.file_util import get_fs
-from bytelatent.distributed import get_global_rank, setup_torch_distributed, DistributedArgs
+from bytelatent.distributed import (
+    DistributedArgs,
+    get_global_rank,
+    setup_torch_distributed,
+)
 from bytelatent.model.blt import ByteLatentTransformer
 from bytelatent.tokenizers.abstract_tokenizer import Tokenizer
 from bytelatent.transformer import LMTransformer
@@ -388,10 +392,7 @@ class PackedCausalTransformerGenerator:
         return generation, loglikelihood, greedy
 
 
-def load_consolidated_model_and_tokenizer(
-    consolidated_path,
-    init_distributed=False
-):
+def load_consolidated_model_and_tokenizer(consolidated_path, init_distributed=False):
     if init_distributed:
         distributed_args = DistributedArgs()
         distributed_args.configure_world()

demo.py

@@ -0,0 +1,43 @@
+import os
+
+import torch
+import typer
+
+from bytelatent.distributed import DistributedArgs, setup_torch_distributed
+from bytelatent.generate import load_consolidated_model_and_tokenizer
+from bytelatent.generate_blt import generate_nocache
+from bytelatent.model.blt import ByteLatentTransformer
+from bytelatent.tokenizers.blt_tokenizer import BltTokenizer
+
+
+def main(prompt: str, model_name: str = "blt-1b"):
+    distributed_args = DistributedArgs()
+    distributed_args.configure_world()
+    if not torch.distributed.is_initialized():
+        setup_torch_distributed(distributed_args)
+    checkpoint_path = os.path.join("hf-weights", model_name)
+    print(f"Loading BLT model: {model_name}")
+    model, tokenizer, train_cfg = load_consolidated_model_and_tokenizer(
+        checkpoint_path,
+    )
+    assert isinstance(model, ByteLatentTransformer)
+    assert isinstance(tokenizer, BltTokenizer)
+    patcher_args = train_cfg.data.patcher_args.model_copy(deep=True)
+    patcher_args.realtime_patching = True
+    print("Loading entropy model and patcher")
+    patcher_args.entropy_model_checkpoint_dir = os.path.join(
+        checkpoint_path, "entropy_model"
+    )
+    patcher = patcher_args.build()
+    prompts = [prompt]
+    outputs = generate_nocache(
+        prompts, model=model, tokenizer=tokenizer, patcher=patcher
+    )
+    text_outputs = [tokenizer.decode(t) for t in outputs]
+    for p, t in zip(prompts, text_outputs):
+        print(f'Prompt: "{p}" Completion: "{t}"')
+        print()
+
+
+if __name__ == "__main__":
+    typer.run(main)

download_blt_weights.py

@@ -0,0 +1,15 @@
+import os
+
+import typer
+from huggingface_hub import snapshot_download
+
+
+def main(models: list[str] = ["blt-1b", "blt-7b"]):
+    if not os.path.exists("hf-weights"):
+        os.makedirs("hf-weights")
+    for model in models:
+        snapshot_download(f"facebook/{model}", local_dir=f"hf-weights/{model}")
+
+
+if __name__ == "__main__":
+    typer.run(main)