Support qwen1.5-*-AWQ model inference in turbomind

[lvhan028]

test the following cases

• lmdeploy chat turbomind Qwen/Qwen1.5-7B-Chat-AWQ --model-format awq

[irexyc]

test
lmdeploy chat turbomind Qwen/Qwen1.5-1.8B-Chat-AWQ --model-format awq

[AllentDan]

Tested lmdeploy chat turbomind Qwen/Qwen1.5-4B-Chat-AWQ --model-format awq

[726663676]

@irexyc @AllentDan @lvhan028 您好，请问目前运行qwen1.5的awq量化有支持并行吗？我编译了支持qwen-awq源码，启动lmdeploy后报错，单卡正常运行
单卡测试lmdeploy chat turbomind Qwen/Qwen1.5-1.8B-Chat-AWQ --model-format awq正常

多卡测试：
1.启动qwen1.5-1.8b-awq
报错信息如下：
lmdeploy chat turbomind /root/.cache/modelscope/hub/qwen/Qwen1___5-1___8B-Chat-AWQ --model-format awq --tp 2
lmdeploy/turbomind/deploy/target_model/base.py", line 240, in save_split
assert tensor.shape[split_dim] % tp == 0
AssertionError
2. 启动qwen1.5-32b-awq
lmdeploy chat turbomind /root/.cache/modelscope/hub/qwen/Qwen1___5-32B-Chat-AWQ --model-format awq --tp 2
报错信息如下：
lmdeploy/turbomind/deploy/target_model/base.py", line 223, in export_weight
tm_tensor.copy_from(torch_tensor)
RuntimeError: [TM][ERROR] Assertion fail: /lmdeploy/src/turbomind/python/bind.cpp:294

[AllentDan]

@726663676 试了 lmdeploy chat turbomind Qwen/Qwen1.5-4B-Chat-AWQ --model-format awq --tp 2 没问题

[726663676]

@AllentDan 您好，我刚试了上面命令lmdeploy chat turbomind Qwen/Qwen1.5-4B-Chat-AWQ --model-format awq --tp 2也能正常运行？但是qwen1.5的1.8b和32b的awq模型没法跑通？

下面测试qwen1.5-4B-awq是正常的，其他大小的模型还是出现上面的异常：
CUDA_VISIBLE_DEVICES=0,4 lmdeploy chat turbomind /root/.cache/modelscope/hub/qwen/Qwen1___5-4B-Chat-AWQ --model-format awq --tp 2
...
2024-04-15 03:30:26,884 - lmdeploy - WARNING - get 883 model params
2024-04-15 03:30:27,886 - lmdeploy - WARNING - Input chat template with model_name is None. Forcing to use qwen
[WARNING] gemm_config.in is not found; using default GEMM algo
[WARNING] gemm_config.in is not found; using default GEMM algo
session 1

double enter to end input >>> hello

<|im_start|>system
You are a helpful assistant.<|im_end|>
<|im_start|>user
�hello<|im_end|>
<|im_start|>assistant
2024-04-15 03:30:37,819 - lmdeploy - WARNING - kwargs ignore_eos is deprecated for inference, use GenerationConfig instead.
2024-04-15 03:30:37,819 - lmdeploy - WARNING - kwargs random_seed is deprecated for inference, use GenerationConfig instead.
Hello! How can I help you today? If you have any questions or need assistance, please feel free to ask.

[lvhan028]

@726663676 我也把其他规模的awq模型下载下来看看
抱歉，我之前只验证了 7B-AWQ 的模型