axolotl/src/axolotl/integrations/kd/trainer.py, line77
if num_items_in_batch is None:
    num_items_in_batch = -1


/usr/local/lib/python3.10/dist-packages/trl/scripts/vllm_serve.py, line 67

add after "os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn""
```
os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
import multiprocessing as mp
try:
    mp.set_start_method('spawn', force=True)
    print("spawned")
except RuntimeError:
    pass
```


"/usr/local/lib/python3.10/dist-packages/vllm/model_executor/models/qwen2.py", line 397, in load_weights
```
load_weights -> https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/qwen2.py # vllm==0.9.0
```
"/usr/local/lib/python3.10/dist-packages/vllm/model_executor/layers/linear.py", line 220, in apply
"/usr/local/lib/python3.10/dist-packages/vllm/model_executor/layers/vocab_parallel_embedding.py", line 57, in apply
```
# return torch.matmul(x, layer.weight)
                if x.shape[-1] == layer.weight.shape[-1]:
                    return torch.matmul(x, layer.weight.permute(1, 0))
                else:
                    return torch.matmul(x, layer.weight)
```