rm flash attn

Dockerfile

@@ -15,8 +15,7 @@ COPY . .
 
 # Install Python dependencies
 RUN pip install --upgrade pip && \
-    pip install --no-cache-dir -r requirements.txt && \
-    MAX_JOBS=4 pip install flash-attn --use-pep517 --no-build-isolation
+    pip install --no-cache-dir -r requirements.txt
 
 # Make script executable and move it to PATH
 RUN install -m 755 sft.py /usr/local/bin/sft

configs/Qwen3-4B-Base_a10g-small.yaml

@@ -2,7 +2,6 @@
 model_name_or_path: Qwen/Qwen2.5-1.5B
 model_revision: main
 torch_dtype: bfloat16
-attn_implementation: flash_attention_2
 
 # Data training arguments
 dataset_name: trl-lib/tldr