Instructions to use tiny-random/glm-4.6v with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tiny-random/glm-4.6v with Transformers:

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

pipe = pipeline("image-text-to-text", model="tiny-random/glm-4.6v")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("tiny-random/glm-4.6v")
model = AutoModelForMultimodalLM.from_pretrained("tiny-random/glm-4.6v", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.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(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use tiny-random/glm-4.6v with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tiny-random/glm-4.6v"
# 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-4.6v",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/tiny-random/glm-4.6v

SGLang

How to use tiny-random/glm-4.6v 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-4.6v" \
    --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-4.6v",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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-4.6v" \
        --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-4.6v",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use tiny-random/glm-4.6v with Docker Model Runner:
```
docker model run hf.co/tiny-random/glm-4.6v
```

glm-4.6v / README.md

yujiepan

Upload folder using huggingface_hub

d847a87 verified 8 months ago

preview code

Raw

History Blame Contribute Delete

8.06 kB

	---
	library_name: transformers
	base_model:
	- zai-org/GLM-4.6V
	---

	This tiny model is intended for debugging. It is randomly initialized using the configuration adapted from [zai-org/GLM-4.6V](https://huggingface.co/zai-org/GLM-4.6V).

	### Example usage:

	```python
	import torch
	from transformers import AutoProcessor, Glm4vMoeForConditionalGeneration

	model_id = "tiny-random/glm-4.6v"
	messages = [
	{
	"role": "user",
	"content": [
	{
	"type": "image",
	"url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"
	},
	{
	"type": "text",
	"text": "describe this image"
	}
	],
	}
	]
	processor = AutoProcessor.from_pretrained(model_id)
	model = Glm4vMoeForConditionalGeneration.from_pretrained(
	model_id,
	torch_dtype=torch.bfloat16,
	device_map="cuda",
	)
	inputs = processor.apply_chat_template(
	messages,
	tokenize=True,
	add_generation_prompt=True,
	return_dict=True,
	return_tensors="pt"
	).to(model.device)
	inputs.pop("token_type_ids", None)
	generated_ids = model.generate(**inputs, max_new_tokens=16)
	output_text = processor.decode(
	generated_ids[0][inputs["input_ids"].shape[1]:], skip_special_tokens=False)
	print(output_text)
	```

	### Codes to create this repo:

	```python
	import json
	from pathlib import Path

	import accelerate
	import torch
	from huggingface_hub import file_exists, hf_hub_download
	from transformers import (
	AutoConfig,
	AutoModelForCausalLM,
	AutoProcessor,
	GenerationConfig,
	Glm4vForConditionalGeneration,
	Glm4vMoeForConditionalGeneration,
	set_seed,
	)
	from transformers.models.glm4v_moe.modeling_glm4v_moe import Glm4vMoeTextTopkRouter

	source_model_id = "zai-org/GLM-4.6V"
	save_folder = "/tmp/tiny-random/glm-4.6v"
	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['text_config'].update({
	"hidden_size": 8,
	"head_dim": 32,
	"intermediate_size": 64,
	"first_k_dense_replace": 1,
	"moe_intermediate_size": 64,
	"num_attention_heads": 8,
	"num_key_value_heads": 4,
	"num_hidden_layers": 2, # one dense, one moe
	"tie_word_embeddings": True,
	})
	config_json['text_config']['rope_parameters']['mrope_section'] = [2, 2, 4]
	config_json['vision_config']['hidden_size'] = 64
	config_json['vision_config']['depth'] = 2
	config_json['vision_config']['num_heads'] = 2
	config_json['vision_config']['intermediate_size'] = 64
	config_json['vision_config']['out_hidden_size'] = config_json['text_config']['hidden_size']

	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 = Glm4vMoeForConditionalGeneration(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,
	)
	set_seed(42)
	model = model.cpu() # cpu is more stable for random initialization across machines
	num_params = sum(p.numel() for p in model.parameters())
	with torch.no_grad():
	for name, p in sorted(model.named_parameters()):
	torch.nn.init.normal_(p, 0, 0.1)
	print(name, p.shape, p.dtype, p.device,
	f'{p.numel() / num_params * 100: .2f}%')
	for _, m in sorted(model.named_modules()):
	if isinstance(m, Glm4vMoeTextTopkRouter):
	assert 'e_score_correction_bias' in m.state_dict()
	torch.nn.init.normal_(m.e_score_correction_bias, 0, 1)
	model.save_pretrained(save_folder)
	print(model)
	```

	### Printing the model:

	```text
	Glm4vMoeForConditionalGeneration(
	(model): Glm4vMoeModel(
	(visual): Glm4vMoeVisionModel(
	(embeddings): Glm4vMoeVisionEmbeddings(
	(position_embedding): Embedding(576, 64)
	)
	(patch_embed): Glm4vMoeVisionPatchEmbed(
	(proj): Conv3d(3, 64, kernel_size=(2, 14, 14), stride=(2, 14, 14))
	)
	(rotary_pos_emb): Glm4vMoeVisionRotaryEmbedding()
	(blocks): ModuleList(
	(0-1): 2 x Glm4vMoeVisionBlock(
	(norm1): Glm4vMoeRMSNorm((64,), eps=1e-05)
	(norm2): Glm4vMoeRMSNorm((64,), eps=1e-05)
	(attn): Glm4vMoeVisionAttention(
	(qkv): Linear(in_features=64, out_features=192, bias=False)
	(proj): Linear(in_features=64, out_features=64, bias=False)
	)
	(mlp): Glm4vMoeisionMlp(
	(gate_proj): Linear(in_features=64, out_features=8, bias=False)
	(up_proj): Linear(in_features=64, out_features=8, bias=False)
	(down_proj): Linear(in_features=8, out_features=64, bias=False)
	(act_fn): SiLUActivation()
	)
	)
	)
	(merger): Glm4vMoeVisionPatchMerger(
	(proj): Linear(in_features=8, out_features=8, bias=False)
	(post_projection_norm): LayerNorm((8,), eps=1e-05, elementwise_affine=True)
	(gate_proj): Linear(in_features=8, out_features=64, bias=False)
	(up_proj): Linear(in_features=8, out_features=64, bias=False)
	(down_proj): Linear(in_features=64, out_features=8, bias=False)
	(act1): GELU(approximate='none')
	(act_fn): SiLUActivation()
	)
	(post_conv_layernorm): Glm4vMoeRMSNorm((64,), eps=1e-05)
	(downsample): Conv2d(64, 8, kernel_size=(2, 2), stride=(2, 2))
	(post_layernorm): Glm4vMoeRMSNorm((64,), eps=1e-05)
	)
	(language_model): Glm4vMoeTextModel(
	(embed_tokens): Embedding(151552, 8, padding_idx=151329)
	(layers): ModuleList(
	(0): Glm4vMoeTextDecoderLayer(
	(self_attn): Glm4vMoeTextAttention(
	(q_proj): Linear(in_features=8, out_features=256, bias=True)
	(k_proj): Linear(in_features=8, out_features=128, bias=True)
	(v_proj): Linear(in_features=8, out_features=128, bias=True)
	(o_proj): Linear(in_features=256, out_features=8, bias=False)
	)
	(mlp): Glm4vMoeTextMLP(
	(gate_proj): Linear(in_features=8, out_features=64, bias=False)
	(up_proj): Linear(in_features=8, out_features=64, bias=False)
	(down_proj): Linear(in_features=64, out_features=8, bias=False)
	(act_fn): SiLUActivation()
	)
	(input_layernorm): Glm4vMoeTextRMSNorm((8,), eps=1e-05)
	(post_attention_layernorm): Glm4vMoeTextRMSNorm((8,), eps=1e-05)
	)
	(1): Glm4vMoeTextDecoderLayer(
	(self_attn): Glm4vMoeTextAttention(
	(q_proj): Linear(in_features=8, out_features=256, bias=True)
	(k_proj): Linear(in_features=8, out_features=128, bias=True)
	(v_proj): Linear(in_features=8, out_features=128, bias=True)
	(o_proj): Linear(in_features=256, out_features=8, bias=False)
	)
	(mlp): Glm4vMoeTextMoE(
	(experts): Glm4vMoeTextNaiveMoe(
	(act_fn): SiLUActivation()
	)
	(gate): Glm4vMoeTextTopkRouter()
	(shared_experts): Glm4vMoeTextMLP(
	(gate_proj): Linear(in_features=8, out_features=64, bias=False)
	(up_proj): Linear(in_features=8, out_features=64, bias=False)
	(down_proj): Linear(in_features=64, out_features=8, bias=False)
	(act_fn): SiLUActivation()
	)
	)
	(input_layernorm): Glm4vMoeTextRMSNorm((8,), eps=1e-05)
	(post_attention_layernorm): Glm4vMoeTextRMSNorm((8,), eps=1e-05)
	)
	)
	(norm): Glm4vMoeRMSNorm((8,), eps=1e-05)
	(rotary_emb): Glm4vMoeTextRotaryEmbedding()
	)
	)
	(lm_head): Linear(in_features=8, out_features=151552, bias=False)
	)
	```