Instructions to use SparseLLM/ProSparse-MiniCPM-1B-sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SparseLLM/ProSparse-MiniCPM-1B-sft with Transformers:

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

pipe = pipeline("text-generation", model="SparseLLM/ProSparse-MiniCPM-1B-sft", trust_remote_code=True)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("SparseLLM/ProSparse-MiniCPM-1B-sft", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use SparseLLM/ProSparse-MiniCPM-1B-sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SparseLLM/ProSparse-MiniCPM-1B-sft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SparseLLM/ProSparse-MiniCPM-1B-sft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/SparseLLM/ProSparse-MiniCPM-1B-sft

SGLang

How to use SparseLLM/ProSparse-MiniCPM-1B-sft 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 "SparseLLM/ProSparse-MiniCPM-1B-sft" \
    --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": "SparseLLM/ProSparse-MiniCPM-1B-sft",
		"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 "SparseLLM/ProSparse-MiniCPM-1B-sft" \
        --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": "SparseLLM/ProSparse-MiniCPM-1B-sft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use SparseLLM/ProSparse-MiniCPM-1B-sft with Docker Model Runner:
```
docker model run hf.co/SparseLLM/ProSparse-MiniCPM-1B-sft
```

Raincleared commited on May 28, 2024

Commit

efca936

verified ·

1 Parent(s): 21b27c5

Update configuration_minicpm.py

Browse files

Files changed (1) hide show

configuration_minicpm.py +6 -2

configuration_minicpm.py CHANGED Viewed

@@ -58,8 +58,10 @@ class MiniCPMConfig(PretrainedConfig):
             by meanpooling all the original heads within that group. For more details checkout [this
             paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
             `num_attention_heads`.
-        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
             The non-linear activation function (function or string) in the decoder.
         max_position_embeddings (`int`, *optional*, defaults to 2048):
             The maximum sequence length that this model might ever be used with. MiniCPM 1 supports up to 2048 tokens,
             MiniCPM 2 up to 4096, CodeMiniCPM up to 16384.
@@ -122,7 +124,8 @@ class MiniCPMConfig(PretrainedConfig):
         num_hidden_layers=32,
         num_attention_heads=32,
         num_key_value_heads=None,
-        hidden_act="silu",
         max_position_embeddings=2048,
         initializer_range=0.02,
         rms_norm_eps=1e-6,
@@ -154,6 +157,7 @@ class MiniCPMConfig(PretrainedConfig):
         self.num_key_value_heads = num_key_value_heads
         self.hidden_act = hidden_act
         self.initializer_range = initializer_range
         self.rms_norm_eps = rms_norm_eps
         self.pretraining_tp = pretraining_tp

             by meanpooling all the original heads within that group. For more details checkout [this
             paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
             `num_attention_heads`.
+        hidden_act (`str` or `function`, *optional*, defaults to `"relu"`):
             The non-linear activation function (function or string) in the decoder.
+        hidden_act_param (`float`, *optional*, defaults to 0.):
+            The bias for shiftrelu or threshold for fatrelu.
         max_position_embeddings (`int`, *optional*, defaults to 2048):
             The maximum sequence length that this model might ever be used with. MiniCPM 1 supports up to 2048 tokens,
             MiniCPM 2 up to 4096, CodeMiniCPM up to 16384.
         num_hidden_layers=32,
         num_attention_heads=32,
         num_key_value_heads=None,
+        hidden_act="relu",
+        hidden_act_param=0.,
         max_position_embeddings=2048,
         initializer_range=0.02,
         rms_norm_eps=1e-6,
         self.num_key_value_heads = num_key_value_heads
         self.hidden_act = hidden_act
+        self.hidden_act_param = hidden_act_param
         self.initializer_range = initializer_range
         self.rms_norm_eps = rms_norm_eps
         self.pretraining_tp = pretraining_tp