Instructions to use internlm/internlm-xcomposer2-7b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use internlm/internlm-xcomposer2-7b with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="internlm/internlm-xcomposer2-7b", trust_remote_code=True)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("internlm/internlm-xcomposer2-7b", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use internlm/internlm-xcomposer2-7b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "internlm/internlm-xcomposer2-7b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "internlm/internlm-xcomposer2-7b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/internlm/internlm-xcomposer2-7b

SGLang

How to use internlm/internlm-xcomposer2-7b with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "internlm/internlm-xcomposer2-7b" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "internlm/internlm-xcomposer2-7b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "internlm/internlm-xcomposer2-7b" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "internlm/internlm-xcomposer2-7b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use internlm/internlm-xcomposer2-7b with Docker Model Runner:
```
docker model run hf.co/internlm/internlm-xcomposer2-7b
```

fix(internlm): Prevent errors by padding the dimensions of wrap tokens.

by yun - opened Mar 11, 2024

base: refs/heads/main

←

from: refs/pr/2

Discussion Files changed

-9

Files changed (1) hide show

modeling_internlm_xcomposer2.py +9 -9

modeling_internlm_xcomposer2.py CHANGED Viewed

@@ -258,15 +258,15 @@ class InternLMXComposer2ForCausalLM(InternLM2PreTrainedModel):
             wrap_target = wrap_target[:, :self.max_length].to(self.device)
             wrap_im_mask = wrap_im_mask[:, :self.max_length].to(self.device)
-            wrap_embeds_list.append(wrap_embeds)
-            wrap_atts_list.append(wrap_atts)
-            wrap_target_list.append(wrap_target)
-            wrap_im_mask_list.append(wrap_im_mask)
-        wrap_embeds = torch.cat(wrap_embeds_list)
-        wrap_atts = torch.cat(wrap_atts_list)
-        wrap_target = torch.cat(wrap_target_list)
-        wrap_im_mask = torch.cat(wrap_im_mask_list)
         return wrap_embeds, wrap_atts, wrap_target, wrap_im_mask
     def mask_human_targets(self, input_ids, pure=False):

             wrap_target = wrap_target[:, :self.max_length].to(self.device)
             wrap_im_mask = wrap_im_mask[:, :self.max_length].to(self.device)
+            wrap_embeds_list.append(wrap_embeds.squeeze(0))
+            wrap_atts_list.append(wrap_atts.squeeze(0))
+            wrap_target_list.append(wrap_target.squeeze(0))
+            wrap_im_mask_list.append(wrap_im_mask.squeeze(0))
+        wrap_embeds = torch.nn.utils.rnn.pad_sequence(wrap_embeds_list, batch_first=True, padding_value=self.tokenizer._pad_token_type_id)
+        wrap_atts = torch.nn.utils.rnn.pad_sequence(wrap_atts_list, batch_first=True, padding_value=self.tokenizer._pad_token_type_id)
+        wrap_target = torch.nn.utils.rnn.pad_sequence(wrap_target_list, batch_first=True, padding_value=self.tokenizer._pad_token_type_id)
+        wrap_im_mask = torch.nn.utils.rnn.pad_sequence(wrap_im_mask_list, batch_first=True, padding_value=self.tokenizer._pad_token_type_id)
         return wrap_embeds, wrap_atts, wrap_target, wrap_im_mask
     def mask_human_targets(self, input_ids, pure=False):