Instructions to use Jon-Nielsen/GLM-4.7-REAP-30-W4A16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Jon-Nielsen/GLM-4.7-REAP-30-W4A16 with Transformers:

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

pipe = pipeline("text-generation", model="Jon-Nielsen/GLM-4.7-REAP-30-W4A16")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Jon-Nielsen/GLM-4.7-REAP-30-W4A16")
model = AutoModelForCausalLM.from_pretrained("Jon-Nielsen/GLM-4.7-REAP-30-W4A16", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Jon-Nielsen/GLM-4.7-REAP-30-W4A16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Jon-Nielsen/GLM-4.7-REAP-30-W4A16"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Jon-Nielsen/GLM-4.7-REAP-30-W4A16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Jon-Nielsen/GLM-4.7-REAP-30-W4A16

SGLang

How to use Jon-Nielsen/GLM-4.7-REAP-30-W4A16 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 "Jon-Nielsen/GLM-4.7-REAP-30-W4A16" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Jon-Nielsen/GLM-4.7-REAP-30-W4A16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "Jon-Nielsen/GLM-4.7-REAP-30-W4A16" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Jon-Nielsen/GLM-4.7-REAP-30-W4A16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Jon-Nielsen/GLM-4.7-REAP-30-W4A16 with Docker Model Runner:
```
docker model run hf.co/Jon-Nielsen/GLM-4.7-REAP-30-W4A16
```

GLM-4.7-REAP-30-W4A16 / README.md

Jon-Nielsen

Add model card

3722bc8 verified 8 months ago

preview code

Raw

History Blame Contribute Delete

1.32 kB

metadata

license: apache-2.0
base_model:
  - zai-org/GLM-4.7
  - 0xSero/GLM-4.7-REAP-30
tags:
  - glm-4
  - moe
  - glm4_moe
  - glm
  - glm4
  - reap
  - pruned
  - quantized
  - gptq
  - auto-round
  - 4bit
  - cerebras
  - code
  - function-calling
pipeline_tag: text-generation
language:
  - en
library_name: transformers

GLM-4.7-REAP-30 4-bit GPTQ Quantization

This is a 4-bit AutoRound GPTQ quantization of 0xSero/GLM-4.7-REAP-30, a 30% REAP-pruned version of the original GLM-4.7 MoE model.

Quantized with AutoRound v0.9.4 (bits=4, group_size=128, sym=True) in auto_gptq format.
Model size: ~124 GB (3.8x compression from the unquantized pruned model).
Compatible with vLLM, Transformers + AutoGPTQ, ExLlamaV2, etc.

Example vLLM serving command

(tested on 2x NVIDIA RTX PRO 6000 Blackwell with the following settings)

vllm serve Jon-Nielsen/GLM-4.7-REAP-30-W4A16 \
  --tensor-parallel-size 2 \
  --kv-cache-dtype fp8 \
  --gpu-memory-utilization 0.96 \
  --max-model-len 196608 \
  --max-num-seqs 8 \
  --max-num-batched-tokens 16384 \
  --trust-remote-code \
  --enable-prefix-caching \
  --enable-chunked-prefill \
  --enable-expert-parallel \
  --tool-call-parser=glm47 \
  --reasoning-parser=glm45 \
  --enable-auto-tool-choice \
  --disable-custom-all-reduce