Add models (#18)

* Add models

* Add comments

configs/eval_internlm_7b.py

@@ -0,0 +1,9 @@
+from mmengine.config import read_base
+
+with read_base():
+    # choose a list of datasets
+    from .datasets.collections.base_medium import datasets
+    # choose a model of interest
+    from .models.hf_internlm_7b import models
+    # and output the results in a choosen format
+    from .summarizers.medium import summarizer

configs/models/hf_baichuan_7b.py

@@ -0,0 +1,20 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='baichuan-7b-hf',
+        path="baichuan-inc/baichuan-7B",
+        tokenizer_path='baichuan-inc/baichuan-7B',
+        tokenizer_kwargs=dict(padding_side='left',
+                              truncation_side='left',
+                              trust_remote_code=True,
+                              use_fast=False,),
+        max_out_len=100,
+        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),
+    )
+]

configs/models/hf_chatglm2_6b.py

@@ -0,0 +1,21 @@
+from opencompass.models import HuggingFace
+
+
+models = [
+    dict(
+        type=HuggingFace,
+        abbr='chatglm2-6b-hf',
+        path='THUDM/chatglm2-6b',
+        tokenizer_path='THUDM/chatglm2-6b',
+        tokenizer_kwargs=dict(
+           padding_side='left',
+           truncation_side='left',
+           trust_remote_code=True,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(trust_remote_code=True, device_map='auto', revision='a6d54fac46dff2db65d53416c207a4485ca6bd40'),
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]

configs/models/hf_chatglm_6b.py

@@ -0,0 +1,21 @@
+from opencompass.models import HuggingFace
+
+
+models = [
+    dict(
+        type=HuggingFace,
+        abbr='chatglm-6b-hf',
+        path='THUDM/chatglm-6b',
+        tokenizer_path='THUDM/chatglm-6b',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            trust_remote_code=True,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(trust_remote_code=True, device_map='auto', revision='1d240ba371910e9282298d4592532d7f0f3e9f3e'),
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]

configs/models/hf_falcon_7b.py

@@ -0,0 +1,21 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='falcon-7b-hf',
+        path='tiiuae/falcon-7b',
+        tokenizer_path='tiiuae/falcon-7b',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            trust_remote_code=True,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(trust_remote_code=True, device_map='auto', revision='2f5c3cd4eace6be6c0f12981f377fb35e5bf6ee5'),
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]

configs/models/hf_internlm_7b.py

@@ -0,0 +1,21 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='internlm-7b-hf',
+        path="internlm-7b",
+        tokenizer_path='internlm-7b',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            use_fast=False,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(trust_remote_code=True, device_map='auto'),
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]

configs/models/hf_internlm_chat_7b.py

@@ -0,0 +1,29 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+_meta_template = dict(
+    round=[
+        dict(role='HUMAN', begin='<|User|>:', end='<eoh>\n'),
+        dict(role='BOT', begin='<|Bot|>:', end='<eoa>\n', generate=True),
+    ],
+)
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='internlm-chat-7b-hf',
+        path="internlm-chat-7b",
+        tokenizer_path='internlm-chat-7b',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            use_fast=False,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        meta_template=_meta_template,
+        model_kwargs=dict(trust_remote_code=True, device_map='auto'),
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]

configs/models/hf_llama_13b.py

@@ -0,0 +1,21 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    # LLaMA 13B
+    dict(
+        type=HuggingFaceCausalLM,
+        path="decapoda-research/llama-13b-hf",
+        tokenizer_path='decapoda-research/llama-13b-hf',
+        tokenizer_kwargs=dict(padding_side='left',
+                              truncation_side='left',
+                              use_fast=False,
+                              ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto'),
+        batch_padding=False, # if false, inference with for-loop without batch padding
+        run_cfg=dict(num_gpus=2, num_procs=1),
+    )
+]

configs/models/hf_llama_30b.py

@@ -0,0 +1,22 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    # LLaMA 30B
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='llama-30b-hf',
+        path="decapoda-research/llama-30b-hf",
+        tokenizer_path='decapoda-research/llama-30b-hf',
+        tokenizer_kwargs=dict(padding_side='left',
+                              truncation_side='left',
+                              use_fast=False,
+                              ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto'),
+        batch_padding=False, # if false, inference with for-loop without batch padding
+        run_cfg=dict(num_gpus=4, num_procs=1),
+    )
+]

configs/models/hf_llama_65b.py

@@ -0,0 +1,22 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    # LLaMA 65B
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='llama-65b-hf',
+        path="decapoda-research/llama-65b-hf",
+        tokenizer_path='decapoda-research/llama-65b-hf',
+        tokenizer_kwargs=dict(padding_side='left',
+                              truncation_side='left',
+                              use_fast=False,
+                              ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto'),
+        batch_padding=False, # if false, inference with for-loop without batch padding
+        run_cfg=dict(num_gpus=8, num_procs=1),
+    )
+]

configs/models/hf_llama_7b.py

@@ -5,6 +5,7 @@ models = [
     # LLaMA 7B
     dict(
         type=HuggingFaceCausalLM,
+        abbr='llama-7b-hf',
         path="decapoda-research/llama-7b-hf",
         tokenizer_path='decapoda-research/llama-7b-hf',
         tokenizer_kwargs=dict(padding_side='left',
@@ -15,8 +16,7 @@ models = [
         max_seq_len=2048,
         batch_size=8,
         model_kwargs=dict(device_map='auto'),
-        batch_padding=False, # if false, inference with for-loop without batch padding 
-        run_cfg=dict(num_gpus=2, num_procs=1),
+        batch_padding=False, # if false, inference with for-loop without batch padding
+        run_cfg=dict(num_gpus=1, num_procs=1),
     )
-
 ]

configs/models/hf_moss_moon_003_base.py

@@ -0,0 +1,21 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='moss-moon-003-base-hf',
+        path='fnlp/moss-moon-003-base',
+        tokenizer_path='fnlp/moss-moon-003-base',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            trust_remote_code=True,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(trust_remote_code=True, device_map='auto', revision='5e406ca0ebbdea11cc3b12aa5932995c692568ac'),
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]

configs/models/hf_moss_moon_003_sft.py

@@ -0,0 +1,21 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='moss-moon-003-sft-hf',
+        path='fnlp/moss-moon-003-sft',
+        tokenizer_path='fnlp/moss-moon-003-sft',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            trust_remote_code=True,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(trust_remote_code=True, device_map='auto', revision='7119d446173035561f40977fb9cb999995bb7517'),
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]

configs/models/hf_mpt_7b.py

@@ -0,0 +1,27 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='mpt-7b-hf',
+        path='mosaicml/mpt-7b',
+        tokenizer_path='mosaicml/mpt-7b',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            trust_remote_code=True,
+            use_fast=True
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(
+            device_map='auto',
+            trust_remote_code=True,
+            max_seq_len=4096,
+            revision='68e1a8e0ebb9b30f3c45c1ef6195980f29063ae2',
+        ),
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]

configs/models/hf_mpt_instruct_7b.py

@@ -0,0 +1,27 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='mpt-instruct-7b-hf',
+        path="mosaicml/mpt-7b-instruct",
+        tokenizer_path="mosaicml/mpt-7b-instruct",
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            trust_remote_code=True,
+            use_fast=True
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(
+            device_map='auto',
+            trust_remote_code=True,
+            max_seq_len=4096,
+            revision='68e1a8e0ebb9b30f3c45c1ef6195980f29063ae2',
+        ),
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]

configs/models/hf_tigerbot_7b_base.py

@@ -0,0 +1,21 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='tigerbot-base-7b-hf',
+        path='TigerResearch/tigerbot-7b-base',
+        tokenizer_path='TigerResearch/tigerbot-7b-base',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            trust_remote_code=True,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(trust_remote_code=True, device_map='auto', revision='5f34fd223586e9efb8eb0e3bc667f03581886992'),
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    ),
+]

configs/models/hf_tigerbot_7b_sft.py

@@ -1,17 +1,17 @@
 from opencompass.models import HuggingFaceCausalLM
 
+
 _meta_template = dict(
     round=[
-        dict(role='HUMAN', begin='\n\n### Instruction:\n:'), 
+        dict(role='HUMAN', begin='\n\n### Instruction:\n:'),
         dict(role='BOT', begin='\n\n### Response:\n:', generate=True),
     ],
 )
 
-
 models = [
     dict(
         type=HuggingFaceCausalLM,
-        abbr='TigerBot-SFT',
+        abbr='tigerbot-sft-7b-hf',
         path="TigerResearch/tigerbot-7b-sft",
         tokenizer_path='TigerResearch/tigerbot-7b-sft',
         tokenizer_kwargs=dict(

configs/models/hf_vicuna_v1.3_13b.py

@@ -0,0 +1,22 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='vicuna-13b-hf',
+        path="lmsys/vicuna-13b-v1.3",
+        tokenizer_path='lmsys/vicuna-13b-v1.3',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            use_fast=False,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto'),
+        batch_padding=False, # if false, inference with for-loop without batch padding
+        run_cfg=dict(num_gpus=2, num_procs=1)
+    )
+]

configs/models/hf_vicuna_v1.3_33b.py

@@ -0,0 +1,22 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='vicuna-33b-hf',
+        path="lmsys/vicuna-33b-v1.3",
+        tokenizer_path='lmsys/vicuna-33b-v1.3',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            use_fast=False,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto'),
+        batch_padding=False, # if false, inference with for-loop without batch padding
+        run_cfg=dict(num_gpus=4, num_procs=1)
+    )
+]

configs/models/hf_vicuna_v1.3_7b.py

@@ -0,0 +1,22 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='vicuna-7b-hf',
+        path="lmsys/vicuna-7b-v1.3",
+        tokenizer_path='lmsys/vicuna-7b-v1.3',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            use_fast=False,
+        ),
+        max_out_len=100,
+        max_seq_len=2048,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto'),
+        batch_padding=False, # if false, inference with for-loop without batch padding
+        run_cfg=dict(num_gpus=1, num_procs=1)
+    )
+]

configs/models/hf_wizardlm_7b.py

@@ -0,0 +1,24 @@
+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='wizardlm-7b-hf',
+        path='TheBloke/wizardLM-7B-HF',
+        tokenizer_path='TheBloke/wizardLM-7B-HF',
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            trust_remote_code=True,
+        ),
+        max_out_len=100,
+        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),
+    )
+]