Fix attn impl

README.md

@@ -73,8 +73,6 @@ See full scores [below](#benchmark-table).
 pip install chandra-ocr
 ```
 
-If you're going to use the huggingface method, we also recommend installing [flash attention](https://github.com/Dao-AILab/flash-attention).
-
 ### From Source
 
 ```bash

chandra/model/hf.py

@@ -74,7 +74,7 @@ def load_model():
         "device_map": device_map,
     }
     if settings.TORCH_ATTN:
-        kwargs["attn_implementation"] = settings.TORCH_ATTN
+        kwargs["attn_implementation"] = settings.TORCH_ATTN_IMPLEMENTATION
 
     model = Qwen3VLForConditionalGeneration.from_pretrained(
         settings.MODEL_CHECKPOINT, **kwargs

chandra/scripts/cli.py

@@ -172,7 +172,7 @@ def save_merged_output(
 @click.option(
     "--batch-size",
     type=int,
-    default=None,
+    default=1,
     help="Number of pages to process in a batch.",
 )
 @click.option(
@@ -194,16 +194,6 @@ def main(
     batch_size: int,
     paginate_output: bool,
 ):
-    if method == "hf":
-        click.echo(
-            "When using '--method hf', ensure that the batch size is set correctly.  We will default to batch size of 1."
-        )
-        if batch_size is None:
-            batch_size = 1
-    elif method == "vllm":
-        if batch_size is None:
-            batch_size = 28
-
     click.echo("Chandra CLI - Starting OCR processing")
     click.echo(f"Input: {input_path}")
     click.echo(f"Output: {output_path}")

chandra/scripts/vllm.py

@@ -17,6 +17,8 @@ def main():
         "-v",
         f"{os.path.expanduser('~')}/.cache/huggingface:/root/.cache/huggingface",
         "--env",
+        f"HUGGING_FACE_HUB_TOKEN={os.getenv('HF_TOKEN')}",
+        "--env",
         "VLLM_ATTENTION_BACKEND=TORCH_SDPA",
         "-p",
         "8000:8000",

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "chandra-ocr"
-version = "0.1.5"
+version = "0.1.2"
 description = "OCR model that converts documents to markdown, HTML, or JSON."
 readme = "README.md"
 requires-python = ">=3.10"