chore: import upstream snapshot with attribution

examples/models/deepseek_ocr/infer.py

@@ -0,0 +1,29 @@
+# pip install "transformers==4.46.3" easydict
+import os
+
+os.environ['CUDA_VISIBLE_DEVICES'] = '0'
+# os.environ['SWIFT_DEBUG'] = '1'
+
+if __name__ == '__main__':
+    from swift.infer_engine import InferRequest, RequestConfig, TransformersEngine
+    engine = TransformersEngine('deepseek-ai/DeepSeek-OCR')
+    infer_request = InferRequest(
+        messages=[{
+            'role': 'user',
+            # or
+            'content': '<image>Free OCR.',
+            # "content": '<image><|grounding|>Convert the document to markdown.',
+        }],
+        images=['https://modelscope-open.oss-cn-hangzhou.aliyuncs.com/images/ocr.png'])
+    request_config = RequestConfig(max_tokens=512, temperature=0)
+    resp_list = engine.infer([infer_request], request_config=request_config)
+    response = resp_list[0].choices[0].message.content
+
+    # use stream
+    request_config = RequestConfig(max_tokens=512, temperature=0, stream=True)
+    gen_list = engine.infer([infer_request], request_config=request_config)
+    for chunk in gen_list[0]:
+        if chunk is None:
+            continue
+        print(chunk.choices[0].delta.content, end='', flush=True)
+    print()

examples/models/deepseek_ocr/train.sh

@@ -0,0 +1,30 @@
+# 24GiB
+pip install "transformers==4.46.3" easydict "peft==0.15.2"
+
+CUDA_VISIBLE_DEVICES=0 \
+swift sft \
+    --model deepseek-ai/DeepSeek-OCR \
+    --dataset 'AI-ModelScope/LaTeX_OCR:human_handwrite#20000' \
+    --load_from_cache_file true \
+    --split_dataset_ratio 0.01 \
+    --tuner_type lora \
+    --torch_dtype bfloat16 \
+    --num_train_epochs 1 \
+    --per_device_train_batch_size 1 \
+    --per_device_eval_batch_size 1 \
+    --learning_rate 1e-4 \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --target_modules all-linear \
+    --freeze_vit true \
+    --freeze_aligner true \
+    --gradient_accumulation_steps 16 \
+    --eval_steps 50 \
+    --save_steps 50 \
+    --save_total_limit 2 \
+    --logging_steps 5 \
+    --max_length 4096 \
+    --output_dir output \
+    --warmup_ratio 0.05 \
+    --dataset_num_proc 4 \
+    --dataloader_num_workers 4