updated local_model settings

configs/eval_oreilly_others.py

@@ -0,0 +1,59 @@
+from mmengine.config import read_base
+from opencompass.partitioners import SizePartitioner
+from opencompass.runners import LocalRunner
+from opencompass.tasks import OpenICLInferTask, OpenICLEvalTask
+
+with read_base():
+    # Datasets
+    from .datasets.oreilly.oreilly_test_gen import oreilly_datasets as oreilly
+    # Models
+    from .models.gpt_4_peiqi import models as gpt_4
+    from .models.gpt_3dot5_turbo_peiqi import models as chatgpt_3dot5_turbo
+    from .local_models.chatglm2_6b import models as chatglm2_6b
+    from .local_models.internlm_chat_7b import models as internlm_chat_7b
+    from .local_models.xverse_13b import models as xverse_13b
+    from .local_models.qwen_chat_7b import models as qwen_chat_7b
+    from .local_models.llama2_7b_chat import models as llama2_chat_7b
+    from .local_models.llama2_13b_chat import models as llama2_chat_13b
+    from .local_models.baichuan_13b_chat import models as baichuan_13b_chat
+    from .models.hf_llama2_7b import models as hf_llama2_7b
+    from .models.hf_chatglm2_6b import models as hf_chatglm2_6b
+    from .models.hf_internlm_chat_7b import models as hf_internlm_chat_7b
+
+
+datasets = [
+    *oreilly
+    ]
+
+from opencompass.models import HuggingFaceCausalLM
+
+
+
+models = [ 
+    # *gpt_4,                 # demo pass
+    # *chatgpt_3dot5_turbo,   # demo pass
+    # *chatglm2_6b,           # demo pass
+    *internlm_chat_7b,      # demo pass?
+    # *xverse_13b,            # demo pass?
+    *qwen_chat_7b,        
+    # *baichuan_13b_chat,
+    *llama2_chat_13b,    
+    # *llama2_chat_7b,
+    # *hf_llama2_7b,
+    # *hf_chatglm2_6b,
+    # *hf_internlm_chat_7b,
+        ]
+
+infer = dict(
+    partitioner=dict(
+        type=SizePartitioner,
+        # max_task_size = 100,
+        # gen_task_coef = 1,
+    ),
+    runner=dict(
+        type=LocalRunner,
+        max_num_workers=3,
+        max_workers_per_gpu=1,
+        task=dict(type=OpenICLInferTask),
+    ),
+)

configs/eval_test_model.py

@@ -14,6 +14,7 @@
     from .local_models.xverse_13b import models as xverse_13b
     from .local_models.qwen_chat_7b import models as qwen_chat_7b
     from .local_models.llama2_7b_chat import models as llama2_chat_7b
+    from .local_models.baichuan_7b import models as baichuan_7b
     from .models.hf_llama2_7b import models as hf_llama2_7b
     from .models.hf_chatglm2_6b import models as hf_chatglm2_6b
     from .models.hf_internlm_chat_7b import models as hf_internlm_chat_7b
@@ -30,25 +31,26 @@
 models = [ 
     # *gpt_4,                 # demo pass
     # *chatgpt_3dot5_turbo,   # demo pass
+    # *baichuan_7b, 
     # *chatglm2_6b,           # demo pass
     # *internlm_chat_7b,      # demo pass?
-    # *xverse_13b,            # demo pass?
-    # *qwen_chat_7b,            
+    *xverse_13b,            # demo pass?
+    *qwen_chat_7b,            
     *llama2_chat_7b,
     # *hf_llama2_7b,
     # *hf_chatglm2_6b,
     # *hf_internlm_chat_7b,
         ]
 
-# infer = dict(
-#     partitioner=dict(
-#         type=SizePartitioner,
-#         max_task_size = 100,
-#         gen_task_coef = 1,
-#     ),
-#     runner=dict(
-#         type=LocalRunner,
-#         max_num_workers=4,
-#         task=dict(type=OpenICLInferTask),
-#     ),
-# )
+infer = dict(
+    partitioner=dict(
+        type=SizePartitioner,
+        # max_task_size = 100,
+        # gen_task_coef = 1,
+    ),
+    runner=dict(
+        type=LocalRunner,
+        max_workers_per_gpu=1,
+        task=dict(type=OpenICLInferTask),
+    ),
+)

configs/local_models/baichuan_13b_base.py

@@ -15,6 +15,6 @@
         max_seq_len=2048,
         batch_size=8,
         model_kwargs=dict(device_map='auto', trust_remote_code=True),
-        run_cfg=dict(num_gpus=1, num_procs=1),
+        run_cfg=dict(num_gpus=2, num_procs=1),
     )
 ]

configs/local_models/baichuan_13b_chat.py

@@ -1,5 +1,11 @@
 from opencompass.models import HuggingFace, HuggingFaceCausalLM
 
+_meta_template = dict(
+    round=[
+        dict(role='HUMAN', begin='<s> ', end='</s>\n'),
+        dict(role='BOT', begin='', end='', generate=True),
+    ],
+)
 
 models = [
     dict(
@@ -14,7 +20,8 @@
         max_out_len=100,
         max_seq_len=2048,
         batch_size=8,
+        meta_template=_meta_template, 
         model_kwargs=dict(device_map='auto', trust_remote_code=True),
-        run_cfg=dict(num_gpus=1, num_procs=1),
+        run_cfg=dict(num_gpus=2, num_procs=1),
     )
 ]

configs/local_models/internlm_chat_7b.py

@@ -1,5 +1,11 @@
 from opencompass.models import HuggingFace, HuggingFaceCausalLM
 
+_meta_template = dict(
+    round=[
+        dict(role='HUMAN', begin='<|User|>:', end='<eoh>\n'),
+        dict(role='BOT', begin='<|Bot|>:', end='', generate=True),
+    ],
+)
 
 models = [
     dict(
@@ -14,10 +20,11 @@
         max_out_len=100,
         max_seq_len=2048,
         batch_size=8,
+        meta_template=_meta_template,
         model_kwargs=dict(device_map='auto', trust_remote_code=True),
         # generate_kwargs=dict(
         #     temperature=0
         # ), 
-        run_cfg=dict(num_gpus=1, num_procs=1),
+        run_cfg=dict(num_gpus=2, num_procs=1),
     )
 ]

configs/local_models/llama2_13b_chat.py

@@ -1,4 +1,4 @@
-from opencompass.models import Llama2Chat
+from opencompass.models import Llama2Chat, HuggingFaceCausalLM
 
 # Please follow the instruction in the Meta AI website https://github.com/facebookresearch/llama
 # and download the LLaMA-2-Chat model and tokenizer to the path './models/llama2/llama/'.
@@ -9,7 +9,7 @@
 # cd llama
 # pip install -e .
 
-api_meta_template = dict(
+_meta_template = dict(
     round=[
         dict(role="HUMAN", api_role="HUMAN"),
         dict(role="BOT", api_role="BOT", generate=True),
@@ -18,14 +18,16 @@
 
 models = [
     dict(
-        abbr="llama-2-13b-chat",
-        type=Llama2Chat,
-        path="./models/llama2/llama/llama-2-13b-chat/",
-        tokenizer_path="./models/llama2/llama/tokenizer.model",
-        meta_template=api_meta_template,
+        abbr="llama-2-7b-chat",
+        # type=Llama2Chat,
+        type=HuggingFaceCausalLM, 
+        path="/mnt/mfs/opsgpt/models/llama-hf/Llama-2-13b-chat-hf/",
+        tokenizer_path="/mnt/mfs/opsgpt/models/llama-hf/Llama-2-13b-chat-hf/",
+        meta_template=_meta_template,
         max_out_len=100,
         max_seq_len=2048,
         batch_size=16,
-        run_cfg=dict(num_gpus=2, num_procs=2),
+        model_kwargs=dict(device_map='auto', trust_remote_code=True),
+        run_cfg=dict(num_gpus=2, num_procs=1),
     ),
 ]

configs/local_models/qwen_chat_7b.py

@@ -1,5 +1,11 @@
 from opencompass.models import HuggingFace, HuggingFaceCausalLM
 
+_meta_template = dict(
+    round=[
+        dict(role='HUMAN', begin='<|im_start|> ', end='<|im_end|><|endoftext|>\n'),
+        dict(role='BOT', begin='<|im_start|> ', end='', generate=True),
+    ],
+)
 
 models = [
     dict(
@@ -14,6 +20,7 @@
         max_out_len=100,
         max_seq_len=2048,
         batch_size=8,
+        meta_template=_meta_template, 
         model_kwargs=dict(device_map='auto', trust_remote_code=True),
         # generate_kwargs=dict(
         #     temperature=0

configs/local_models/xverse_13b.py

@@ -18,6 +18,6 @@
         # generate_kwargs=dict(
         #     temperature=0
         # ), 
-        run_cfg=dict(num_gpus=1, num_procs=1),
+        run_cfg=dict(num_gpus=2, num_procs=1),
     )
 ]