Instructions to use tiny-random/glm-4.7-flash with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tiny-random/glm-4.7-flash with Transformers:

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

pipe = pipeline("text-generation", model="tiny-random/glm-4.7-flash")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tiny-random/glm-4.7-flash")
model = AutoModelForCausalLM.from_pretrained("tiny-random/glm-4.7-flash", 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 tiny-random/glm-4.7-flash with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/tiny-random/glm-4.7-flash

SGLang

How to use tiny-random/glm-4.7-flash 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 "tiny-random/glm-4.7-flash" \
    --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": "tiny-random/glm-4.7-flash",
		"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 "tiny-random/glm-4.7-flash" \
        --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": "tiny-random/glm-4.7-flash",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tiny-random/glm-4.7-flash with Docker Model Runner:
```
docker model run hf.co/tiny-random/glm-4.7-flash
```

glm-4.7-flash / README.md

yujiepan

Upload folder using huggingface_hub

587e67b verified 6 months ago

preview code

Raw

History Blame Contribute Delete

8.97 kB

	---
	library_name: transformers
	base_model:
	- zai-org/GLM-4.7-Flash
	---

	This tiny model is intended for debugging. It is randomly initialized using the configuration adapted from [zai-org/GLM-4.7-Flash](https://huggingface.co/zai-org/GLM-4.7-Flash).

	### Example usage:

	- vLLM

	```bash
	# Multi-token prediction is supported
	model_id=tiny-random/glm-4.7-flash
	vllm serve $model_id \
	--tensor-parallel-size 2 \
	--speculative-config.method mtp \
	--speculative-config.num_speculative_tokens 1 \
	--tool-call-parser glm47 \
	--reasoning-parser glm45 \
	--enable-auto-tool-choice
	```

	- SGLang

	```bash
	# Multi-token prediction is supported
	model_id=tiny-random/glm-4.7-flash
	python3 -m sglang.launch_server --model-path $model_id --tp-size 2 \
	--tool-call-parser glm47 \
	--reasoning-parser glm45 \
	--speculative-algorithm EAGLE \
	--speculative-num-steps 3 \
	--speculative-eagle-topk 1 \
	--speculative-num-draft-tokens 4
	```

	- Transformers

	```python
	import torch
	from transformers import AutoModelForCausalLM, AutoTokenizer

	# Load model and tokenizer
	model_id = "tiny-random/glm-4.7-flash"
	messages = [{"role": "user", "content": "hello"}]
	tokenizer = AutoTokenizer.from_pretrained(model_id)
	inputs = tokenizer.apply_chat_template(
	messages,
	tokenize=True,
	add_generation_prompt=True,
	return_dict=True,
	return_tensors="pt",
	)
	model = AutoModelForCausalLM.from_pretrained(
	pretrained_model_name_or_path=model_id,
	torch_dtype=torch.bfloat16,
	device_map="cuda",
	)
	inputs = inputs.to(model.device)
	generated_ids = model.generate(
	**inputs, max_new_tokens=32, do_sample=False)
	output_text = tokenizer.decode(
	generated_ids[0][inputs.input_ids.shape[1]:])
	print(output_text)
	```

	### Codes to create this repo:

	```python
	import json
	from copy import deepcopy
	from pathlib import Path

	import accelerate
	import torch
	import torch.nn as nn
	from huggingface_hub import file_exists, hf_hub_download
	from transformers import (
	AutoConfig,
	AutoModelForCausalLM,
	AutoProcessor,
	GenerationConfig,
	set_seed,
	)

	source_model_id = "zai-org/GLM-4.7-Flash"
	save_folder = "/tmp/tiny-random/glm-4.7-flash"

	processor = AutoProcessor.from_pretrained(
	source_model_id, trust_remote_code=True)
	processor.save_pretrained(save_folder)

	with open(hf_hub_download(source_model_id, filename='config.json', repo_type='model'), 'r', encoding='utf-8') as f:
	config_json = json.load(f)
	config_json.update({
	'kv_lora_rank': 384,
	'num_key_value_heads': 1,
	'q_lora_rank': 32,
	'qk_nope_head_dim': 64,
	'qk_rope_head_dim': 192,
	'v_head_dim': 64,
	'num_key_value_heads': 4,
	'num_attention_heads': 4,
	})
	config_json['hidden_size'] = 8
	config_json['intermediate_size'] = 32
	config_json['moe_intermediate_size'] = 32
	config_json['num_hidden_layers'] = 2
	config_json['tie_word_embeddings'] = False
	config_json['use_cache'] = True
	with open(f"{save_folder}/config.json", "w", encoding='utf-8') as f:
	json.dump(config_json, f, indent=2)

	config = AutoConfig.from_pretrained(
	save_folder,
	trust_remote_code=True,
	)
	print(config)
	torch.set_default_dtype(torch.bfloat16)
	model = AutoModelForCausalLM.from_config(config)
	torch.set_default_dtype(torch.float32)
	if file_exists(filename="generation_config.json", repo_id=source_model_id, repo_type='model'):
	model.generation_config = GenerationConfig.from_pretrained(
	source_model_id, trust_remote_code=True,
	)
	model.generation_config.do_sample = True
	print(model.generation_config)
	model = model.cpu()
	set_seed(42)
	with torch.no_grad():
	for name, p in sorted(model.named_parameters()):
	torch.nn.init.normal_(p, 0, 0.1)
	print(name, p.shape)
	# MTP
	set_seed(42)
	model.model.layers.append(nn.ModuleDict(dict(
	embed_tokens=deepcopy(model.model.embed_tokens),
	shared_head=nn.ModuleDict(dict(
	norm=nn.RMSNorm(config.hidden_size),
	head=deepcopy(model.model.embed_tokens),
	)),
	eh_proj=nn.Linear(config.hidden_size * 2,
	config.hidden_size, bias=False),
	enorm=nn.RMSNorm(config.hidden_size),
	hnorm=nn.RMSNorm(config.hidden_size),
	input_layernorm=nn.RMSNorm(config.hidden_size),
	post_attention_layernorm=nn.RMSNorm(config.hidden_size),
	self_attn=deepcopy(model.model.layers[1].self_attn),
	mlp=deepcopy(model.model.layers[1].mlp),
	)))
	for i in range(1, len(model.model.layers)):
	model.model.layers[i].mlp.gate.e_score_correction_bias = torch.rand_like(
	model.model.layers[i].mlp.gate.e_score_correction_bias).float()
	model.save_pretrained(save_folder)
	print(model)
	```

	### Printing the model:

	```text
	Glm4MoeLiteForCausalLM(
	(model): Glm4MoeLiteModel(
	(embed_tokens): Embedding(154880, 8, padding_idx=154820)
	(layers): ModuleList(
	(0): Glm4MoeLiteDecoderLayer(
	(self_attn): Glm4MoeLiteAttention(
	(q_a_proj): Linear(in_features=8, out_features=32, bias=False)
	(q_a_layernorm): Glm4MoeLiteRMSNorm((32,), eps=1e-06)
	(q_b_proj): Linear(in_features=32, out_features=1024, bias=False)
	(kv_a_proj_with_mqa): Linear(in_features=8, out_features=576, bias=False)
	(kv_a_layernorm): Glm4MoeLiteRMSNorm((384,), eps=1e-06)
	(kv_b_proj): Linear(in_features=384, out_features=512, bias=False)
	(o_proj): Linear(in_features=256, out_features=8, bias=False)
	)
	(mlp): Glm4MoeLiteMLP(
	(gate_proj): Linear(in_features=8, out_features=32, bias=False)
	(up_proj): Linear(in_features=8, out_features=32, bias=False)
	(down_proj): Linear(in_features=32, out_features=8, bias=False)
	(act_fn): SiLUActivation()
	)
	(input_layernorm): Glm4MoeLiteRMSNorm((8,), eps=1e-05)
	(post_attention_layernorm): Glm4MoeLiteRMSNorm((8,), eps=1e-05)
	)
	(1): Glm4MoeLiteDecoderLayer(
	(self_attn): Glm4MoeLiteAttention(
	(q_a_proj): Linear(in_features=8, out_features=32, bias=False)
	(q_a_layernorm): Glm4MoeLiteRMSNorm((32,), eps=1e-06)
	(q_b_proj): Linear(in_features=32, out_features=1024, bias=False)
	(kv_a_proj_with_mqa): Linear(in_features=8, out_features=576, bias=False)
	(kv_a_layernorm): Glm4MoeLiteRMSNorm((384,), eps=1e-06)
	(kv_b_proj): Linear(in_features=384, out_features=512, bias=False)
	(o_proj): Linear(in_features=256, out_features=8, bias=False)
	)
	(mlp): Glm4MoeLiteMoE(
	(experts): Glm4MoeLiteNaiveMoe(
	(act_fn): SiLUActivation()
	)
	(gate): Glm4MoeLiteTopkRouter()
	(shared_experts): Glm4MoeLiteMLP(
	(gate_proj): Linear(in_features=8, out_features=32, bias=False)
	(up_proj): Linear(in_features=8, out_features=32, bias=False)
	(down_proj): Linear(in_features=32, out_features=8, bias=False)
	(act_fn): SiLUActivation()
	)
	)
	(input_layernorm): Glm4MoeLiteRMSNorm((8,), eps=1e-05)
	(post_attention_layernorm): Glm4MoeLiteRMSNorm((8,), eps=1e-05)
	)
	(2): ModuleDict(
	(embed_tokens): Embedding(154880, 8, padding_idx=154820)
	(shared_head): ModuleDict(
	(norm): RMSNorm((8,), eps=None, elementwise_affine=True)
	(head): Embedding(154880, 8, padding_idx=154820)
	)
	(eh_proj): Linear(in_features=16, out_features=8, bias=False)
	(enorm): RMSNorm((8,), eps=None, elementwise_affine=True)
	(hnorm): RMSNorm((8,), eps=None, elementwise_affine=True)
	(input_layernorm): RMSNorm((8,), eps=None, elementwise_affine=True)
	(post_attention_layernorm): RMSNorm((8,), eps=None, elementwise_affine=True)
	(self_attn): Glm4MoeLiteAttention(
	(q_a_proj): Linear(in_features=8, out_features=32, bias=False)
	(q_a_layernorm): Glm4MoeLiteRMSNorm((32,), eps=1e-06)
	(q_b_proj): Linear(in_features=32, out_features=1024, bias=False)
	(kv_a_proj_with_mqa): Linear(in_features=8, out_features=576, bias=False)
	(kv_a_layernorm): Glm4MoeLiteRMSNorm((384,), eps=1e-06)
	(kv_b_proj): Linear(in_features=384, out_features=512, bias=False)
	(o_proj): Linear(in_features=256, out_features=8, bias=False)
	)
	(mlp): Glm4MoeLiteMoE(
	(experts): Glm4MoeLiteNaiveMoe(
	(act_fn): SiLUActivation()
	)
	(gate): Glm4MoeLiteTopkRouter()
	(shared_experts): Glm4MoeLiteMLP(
	(gate_proj): Linear(in_features=8, out_features=32, bias=False)
	(up_proj): Linear(in_features=8, out_features=32, bias=False)
	(down_proj): Linear(in_features=32, out_features=8, bias=False)
	(act_fn): SiLUActivation()
	)
	)
	)
	)
	(norm): Glm4MoeLiteRMSNorm((8,), eps=1e-05)
	(rotary_emb): Glm4MoeLiteRotaryEmbedding()
	)
	(lm_head): Linear(in_features=8, out_features=154880, bias=False)
	)
	```