Instructions to use tiny-random/qwen3.8-fp8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tiny-random/qwen3.8-fp8 with Transformers:

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

pipe = pipeline("image-text-to-text", model="tiny-random/qwen3.8-fp8")
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)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("tiny-random/qwen3.8-fp8")
model = AutoModelForMultimodalLM.from_pretrained("tiny-random/qwen3.8-fp8", 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=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use tiny-random/qwen3.8-fp8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tiny-random/qwen3.8-fp8"
# 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/qwen3.8-fp8",
		"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/qwen3.8-fp8

SGLang

How to use tiny-random/qwen3.8-fp8 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/qwen3.8-fp8" \
    --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/qwen3.8-fp8",
		"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/qwen3.8-fp8" \
        --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/qwen3.8-fp8",
		"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/qwen3.8-fp8 with Docker Model Runner:
```
docker model run hf.co/tiny-random/qwen3.8-fp8
```

qwen3.8-fp8 / README.md

yujiepan

Show verified image/video examples and actual model structures

6d16aea verified 4 days ago

preview code

Raw History Blame Contribute Delete

15.2 kB

---
library_name: transformers
base_model:
- Qwen/Qwen3.8-27B-FP8
---
This tiny model is intended for debugging. It is randomly initialized using the configuration adapted from [Qwen/Qwen3.8-27B-FP8](https://huggingface.co/Qwen/Qwen3.8-27B-FP8).

Structural comparison with `Qwen/Qwen3.8-27B-FP8`:

| Feature | Source | Tiny |
|---|---|---|
| Input modalities | Text, image, video | Same |
| Vocabulary | 248,320 | Same; special-token IDs preserved |
| Backbone layers | 64 | 8 |
| Hidden size | 5,120 | 32 |
| Attention Q/KV heads; head dim | 24/4; 256 | 12/2; 128 |
| Attention cycle | 3 linear + 1 full | Same; two complete cycles |
| Linear-attention key/value heads | 16/48 | 2/6 |
| Linear-attention key/value head dim | 128/128 | Same |
| MLP intermediate size | 17,408 | 512 |
| MTP | 1 full-attention layer; shared embeddings | Same |
| Vision layers / hidden size | 27 / 1,152 | 2 / 64 |
| Vision patch / temporal patch / merge | 16 / 2 / 2 | Same |
| Vision output size | 5,120 | 32 |
| Checkpoint format | FP8 language/MTP projections; BF16 vision/gates; 128x128 blocks | Same FP8/BF16 layout; 32x32 blocks |

| File path | Size |
|------|------|
| model.safetensors | 34.1MB |


### Example usage:

- SGLang (✅ verified)

  - Runs with TP=2, the embedded NEXTN/MTP layer, and CUDA Graph.
  - Verified for text, image-plus-text and video-plus-text on two NVIDIA H20 GPUs;
    see Test environment for dependency versions.
```bash
model_id=tiny-random/qwen3.8-fp8
sglang serve \
  --model-path $model_id \
  --tp-size 2 \
  --fp8-gemm-backend triton \
  --moe-runner-backend triton \
  --speculative-algorithm NEXTN \
  --speculative-num-steps 3 \
  --speculative-eagle-topk 1 \
  --speculative-num-draft-tokens 4 \
  --context-length 4096 \
  --max-total-tokens 4096 \
  --chunked-prefill-size 512 \
  --mem-fraction-static 0.4 \
  --cuda-graph-config '{"decode": {"bs": [1, 2, 4, 8]}}' \
  --disable-custom-all-reduce
```

- Transformers (✅ verified)

```python
import numpy as np
import torch
from PIL import Image
from transformers import AutoProcessor, Qwen3_5ForConditionalGeneration
from transformers.video_utils import VideoMetadata
model_id = "tiny-random/qwen3.8-fp8"
model = Qwen3_5ForConditionalGeneration.from_pretrained(
    model_id, dtype=torch.bfloat16, device_map="cuda", attn_implementation="sdpa",
)
processor = AutoProcessor.from_pretrained(model_id)
image = Image.new("RGB", (64, 64), (220, 30, 30))
# Four predecoded RGB frames, with timestamps supplied through video_metadata.
video = np.stack([np.asarray(Image.new("RGB", (64, 64), color)) for color in
                  [(220, 30, 30), (30, 220, 30), (30, 30, 220), (220, 220, 30)]])
messages = [{"role": "user", "content": [
    {"type": "image", "image": image},
    {"type": "video", "video": video},
    {"type": "text", "text": "Describe the image and the video."},
]}]
inputs = processor.apply_chat_template(
    messages, tokenize=True, add_generation_prompt=True,
    return_dict=True, return_tensors="pt",
    do_sample_frames=False,
    video_metadata=[VideoMetadata(total_num_frames=4, fps=2, frames_indices=[0, 1, 2, 3])],
).to(model.device)
output = model.generate(**inputs, do_sample=False, max_new_tokens=10)
print(processor.decode(output[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True))
```

### Codes to create this repo (✅ verified):

<details><summary>Click to expand</summary>

```python
import json
from copy import deepcopy
from pathlib import Path
from shutil import copyfile
import torch
from huggingface_hub import hf_hub_download, list_repo_files
from safetensors.torch import save_file
from transformers import AutoConfig, AutoModelForImageTextToText, set_seed
source_model_id = "Qwen/Qwen3.8-27B-FP8"
save_folder = "/tmp/tiny-random/qwen38-fp8"
source_revision = '017b9c7af6b5689d5dd426a76e0bc077eb5ca20a'
set_seed(42)
folder = Path(save_folder)
folder.mkdir(parents=True, exist_ok=True)
raw = json.loads(Path(hf_hub_download(source_model_id, "config.json", revision=source_revision)).read_text())
quantization = raw.pop("quantization_config")
quantization["weight_block_size"] = [32, 32]
raw["text_config"].update(
    hidden_size=32, head_dim=128, num_hidden_layers=8,
    layer_types=(["linear_attention"] * 3 + ["full_attention"]) * 2,
    num_attention_heads=12, num_key_value_heads=2,
    linear_num_key_heads=2, linear_num_value_heads=6,
    linear_key_head_dim=128, linear_value_head_dim=128,
    intermediate_size=512,
)
raw["text_config"]["rope_parameters"]["mrope_section"] = [6, 5, 5]
raw["vision_config"].update(depth=2, hidden_size=64, intermediate_size=128, num_heads=2, out_hidden_size=32)
(folder / "config.json").write_text(json.dumps(raw, indent=2))
config = AutoConfig.from_pretrained(folder)
model = AutoModelForImageTextToText.from_config(config, dtype=torch.bfloat16)
backbone = model.model.language_model
text = backbone.config
def norm():
    return type(backbone.norm)(text.hidden_size, text.rms_norm_eps).to(torch.bfloat16)
model.mtp = torch.nn.ModuleDict({
    "pre_fc_norm_embedding": norm(),
    "pre_fc_norm_hidden": norm(),
    "fc": torch.nn.Linear(2 * text.hidden_size, text.hidden_size, bias=False, dtype=torch.bfloat16),
    "layers": torch.nn.ModuleList([deepcopy(backbone.layers[-1])]),
    "norm": norm(),
})
with torch.no_grad():
    for name, param in sorted(model.named_parameters()):
        if name.endswith("A_log"):
            param.zero_()
        elif name.endswith("dt_bias"):
            param.fill_(1)
        elif "norm" in name and name.endswith("weight"):
            param.fill_(1 if "visual" in name or ".linear_attn.norm." in name else 0)
        elif name.endswith("bias"):
            param.zero_()
        else:
            torch.nn.init.normal_(param, mean=0, std=0.02)
state = {name: value.detach().cpu().contiguous() for name, value in model.state_dict().items()}
# Prune source exclusions absent from this architecture before FP8 loading.
names = set(dict(model.named_modules())) | set(dict(model.named_parameters()))
exclusions = [name for name in quantization["modules_to_not_convert"] if name in names]
quantization["modules_to_not_convert"] = exclusions
def pack(name, weight):
    n, k = weight.shape
    bn, bk = quantization["weight_block_size"]
    blocks = weight.float().reshape(n // bn, bn, k // bk, bk)
    scales = (blocks.abs().amax(dim=(1, 3)) / 448).clamp_min(1e-12).to(torch.bfloat16)
    packed = (blocks / scales.float()[:, None, :, None]).clamp(-448, 448)
    state[name] = packed.reshape(n, k).to(torch.float8_e4m3fn).contiguous()
    state[name.removesuffix("weight") + "weight_scale_inv"] = scales.contiguous()
for name, module in model.named_modules():
    excluded = any(name == prefix or name.startswith(prefix + ".") for prefix in exclusions)
    if isinstance(module, torch.nn.Linear) and not excluded:
        pack(name + ".weight", state[name + ".weight"])
save_file(state, folder / "model.safetensors", metadata={"format": "pt"})
raw["quantization_config"] = quantization
(folder / "config.json").write_text(json.dumps(raw, indent=2))
# Preserve tokenizer, processor and generation assets exactly as published.
assets = {
    "tokenizer.json", "tokenizer_config.json", "special_tokens_map.json",
    "vocab.json", "merges.txt", "added_tokens.json", "chat_template.jinja",
    "chat_template.json", "preprocessor_config.json", "processor_config.json",
    "video_preprocessor_config.json", "generation_config.json",
}
for name in sorted(set(list_repo_files(source_model_id, revision=source_revision)) & assets):
    copyfile(hf_hub_download(source_model_id, name, revision=source_revision), folder / name)
```

</details>

### Printing the model:

<details><summary>Click to expand</summary>

```text
Qwen3_5ForConditionalGeneration(
  (model): Qwen3_5Model(
    (visual): Qwen3_5VisionModel(
      (patch_embed): Qwen3_5VisionPatchEmbed(
        (proj): Conv3d(3, 64, kernel_size=(2, 16, 16), stride=(2, 16, 16))
      )
      (pos_embed): Embedding(2304, 64)
      (rotary_pos_emb): Qwen3_5VisionRotaryEmbedding()
      (blocks): ModuleList(
        (0-1): 2 x Qwen3_5VisionBlock(
          (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True, bias=True)
          (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True, bias=True)
          (attn): Qwen3_5VisionAttention(
            (qkv): Linear(in_features=64, out_features=192, bias=True)
            (proj): Linear(in_features=64, out_features=64, bias=True)
          )
          (mlp): Qwen3_5VisionMLP(
            (linear_fc1): Linear(in_features=64, out_features=128, bias=True)
            (linear_fc2): Linear(in_features=128, out_features=64, bias=True)
            (act_fn): GELUTanh()
          )
        )
      )
      (merger): Qwen3_5VisionPatchMerger(
        (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True, bias=True)
        (linear_fc1): Linear(in_features=256, out_features=256, bias=True)
        (act_fn): GELU(approximate='none')
        (linear_fc2): Linear(in_features=256, out_features=32, bias=True)
      )
    )
    (language_model): Qwen3_5TextModel(
      (embed_tokens): Embedding(248320, 32)
      (layers): ModuleList(
        (0-2): 3 x Qwen3_5DecoderLayer(
          (linear_attn): Qwen3_5GatedDeltaNet(
            (conv1d): Conv1d(1280, 1280, kernel_size=(4,), stride=(1,), padding=(3,), groups=1280, bias=False)
            (norm): Qwen3_5RMSNormGated()
            (out_proj): Linear(in_features=768, out_features=32, bias=False)
            (in_proj_qkv): Linear(in_features=32, out_features=1280, bias=False)
            (in_proj_z): Linear(in_features=32, out_features=768, bias=False)
            (in_proj_b): Linear(in_features=32, out_features=6, bias=False)
            (in_proj_a): Linear(in_features=32, out_features=6, bias=False)
          )
          (mlp): Qwen3_5MLP(
            (gate_proj): Linear(in_features=32, out_features=512, bias=False)
            (up_proj): Linear(in_features=32, out_features=512, bias=False)
            (down_proj): Linear(in_features=512, out_features=32, bias=False)
            (act_fn): SiLUActivation()
          )
          (input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
          (post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
        )
        (3): Qwen3_5DecoderLayer(
          (self_attn): Qwen3_5Attention(
            (q_proj): Linear(in_features=32, out_features=3072, bias=False)
            (k_proj): Linear(in_features=32, out_features=256, bias=False)
            (v_proj): Linear(in_features=32, out_features=256, bias=False)
            (o_proj): Linear(in_features=1536, out_features=32, bias=False)
            (q_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
            (k_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
          )
          (mlp): Qwen3_5MLP(
            (gate_proj): Linear(in_features=32, out_features=512, bias=False)
            (up_proj): Linear(in_features=32, out_features=512, bias=False)
            (down_proj): Linear(in_features=512, out_features=32, bias=False)
            (act_fn): SiLUActivation()
          )
          (input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
          (post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
        )
        (4-6): 3 x Qwen3_5DecoderLayer(
          (linear_attn): Qwen3_5GatedDeltaNet(
            (conv1d): Conv1d(1280, 1280, kernel_size=(4,), stride=(1,), padding=(3,), groups=1280, bias=False)
            (norm): Qwen3_5RMSNormGated()
            (out_proj): Linear(in_features=768, out_features=32, bias=False)
            (in_proj_qkv): Linear(in_features=32, out_features=1280, bias=False)
            (in_proj_z): Linear(in_features=32, out_features=768, bias=False)
            (in_proj_b): Linear(in_features=32, out_features=6, bias=False)
            (in_proj_a): Linear(in_features=32, out_features=6, bias=False)
          )
          (mlp): Qwen3_5MLP(
            (gate_proj): Linear(in_features=32, out_features=512, bias=False)
            (up_proj): Linear(in_features=32, out_features=512, bias=False)
            (down_proj): Linear(in_features=512, out_features=32, bias=False)
            (act_fn): SiLUActivation()
          )
          (input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
          (post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
        )
        (7): Qwen3_5DecoderLayer(
          (self_attn): Qwen3_5Attention(
            (q_proj): Linear(in_features=32, out_features=3072, bias=False)
            (k_proj): Linear(in_features=32, out_features=256, bias=False)
            (v_proj): Linear(in_features=32, out_features=256, bias=False)
            (o_proj): Linear(in_features=1536, out_features=32, bias=False)
            (q_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
            (k_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
          )
          (mlp): Qwen3_5MLP(
            (gate_proj): Linear(in_features=32, out_features=512, bias=False)
            (up_proj): Linear(in_features=32, out_features=512, bias=False)
            (down_proj): Linear(in_features=512, out_features=32, bias=False)
            (act_fn): SiLUActivation()
          )
          (input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
          (post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
        )
      )
      (norm): Qwen3_5RMSNorm((32,), eps=1e-06)
      (rotary_emb): Qwen3_5TextRotaryEmbedding()
    )
  )
  (lm_head): Linear(in_features=32, out_features=248320, bias=False)
  (mtp): ModuleDict(
    (pre_fc_norm_embedding): Qwen3_5RMSNorm((32,), eps=1e-06)
    (pre_fc_norm_hidden): Qwen3_5RMSNorm((32,), eps=1e-06)
    (fc): Linear(in_features=64, out_features=32, bias=False)
    (layers): ModuleList(
      (0): Qwen3_5DecoderLayer(
        (self_attn): Qwen3_5Attention(
          (q_proj): Linear(in_features=32, out_features=3072, bias=False)
          (k_proj): Linear(in_features=32, out_features=256, bias=False)
          (v_proj): Linear(in_features=32, out_features=256, bias=False)
          (o_proj): Linear(in_features=1536, out_features=32, bias=False)
          (q_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
          (k_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
        )
        (mlp): Qwen3_5MLP(
          (gate_proj): Linear(in_features=32, out_features=512, bias=False)
          (up_proj): Linear(in_features=32, out_features=512, bias=False)
          (down_proj): Linear(in_features=512, out_features=32, bias=False)
          (act_fn): SiLUActivation()
        )
        (input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
        (post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
      )
    )
    (norm): Qwen3_5RMSNorm((32,), eps=1e-06)
  )
)
```

</details>

### Test environment:

- Pillow: 12.2.0
- accelerate: 1.15.0
- av: 19.0.2
- flashinfer-python: 0.7.2
- huggingface_hub: 1.33.0
- kernels: 0.17.0
- numpy: 2.3.5
- safetensors: 0.8.0
- sglang: 0.5.23.dev384+g9ab74b851
- torch: 2.14.1
- torchcodec: 0.17.0
- transformers: 5.19.0
### Change log:
- 2026-10-23: Initial version.
- 2026-10-23: Reduced all repo files below 50 MB; GPU-validated TP=2, MTP and CUDA Graph.
- 2026-10-23: Creation examples use only public dependencies; source assets are copied unchanged.
This repo is generated by GPT-6.1-Sol

	---
	library_name: transformers
	base_model:
	- Qwen/Qwen3.8-27B-FP8
	---

	This tiny model is intended for debugging. It is randomly initialized using the configuration adapted from [Qwen/Qwen3.8-27B-FP8](https://huggingface.co/Qwen/Qwen3.8-27B-FP8).

	Structural comparison with `Qwen/Qwen3.8-27B-FP8`:

	\| Feature \| Source \| Tiny \|
	\|---\|---\|---\|
	\| Input modalities \| Text, image, video \| Same \|
	\| Vocabulary \| 248,320 \| Same; special-token IDs preserved \|
	\| Backbone layers \| 64 \| 8 \|
	\| Hidden size \| 5,120 \| 32 \|
	\| Attention Q/KV heads; head dim \| 24/4; 256 \| 12/2; 128 \|
	\| Attention cycle \| 3 linear + 1 full \| Same; two complete cycles \|
	\| Linear-attention key/value heads \| 16/48 \| 2/6 \|
	\| Linear-attention key/value head dim \| 128/128 \| Same \|
	\| MLP intermediate size \| 17,408 \| 512 \|
	\| MTP \| 1 full-attention layer; shared embeddings \| Same \|
	\| Vision layers / hidden size \| 27 / 1,152 \| 2 / 64 \|
	\| Vision patch / temporal patch / merge \| 16 / 2 / 2 \| Same \|
	\| Vision output size \| 5,120 \| 32 \|
	\| Checkpoint format \| FP8 language/MTP projections; BF16 vision/gates; 128x128 blocks \| Same FP8/BF16 layout; 32x32 blocks \|

	\| File path \| Size \|
	\|------\|------\|
	\| model.safetensors \| 34.1MB \|


	### Example usage:

	- SGLang (✅ verified)

	- Runs with TP=2, the embedded NEXTN/MTP layer, and CUDA Graph.
	- Verified for text, image-plus-text and video-plus-text on two NVIDIA H20 GPUs;
	see Test environment for dependency versions.

	```bash
	model_id=tiny-random/qwen3.8-fp8
	sglang serve \
	--model-path $model_id \
	--tp-size 2 \
	--fp8-gemm-backend triton \
	--moe-runner-backend triton \
	--speculative-algorithm NEXTN \
	--speculative-num-steps 3 \
	--speculative-eagle-topk 1 \
	--speculative-num-draft-tokens 4 \
	--context-length 4096 \
	--max-total-tokens 4096 \
	--chunked-prefill-size 512 \
	--mem-fraction-static 0.4 \
	--cuda-graph-config '{"decode": {"bs": [1, 2, 4, 8]}}' \
	--disable-custom-all-reduce
	```

	- Transformers (✅ verified)

	```python
	import numpy as np
	import torch
	from PIL import Image
	from transformers import AutoProcessor, Qwen3_5ForConditionalGeneration
	from transformers.video_utils import VideoMetadata

	model_id = "tiny-random/qwen3.8-fp8"
	model = Qwen3_5ForConditionalGeneration.from_pretrained(
	model_id, dtype=torch.bfloat16, device_map="cuda", attn_implementation="sdpa",
	)
	processor = AutoProcessor.from_pretrained(model_id)
	image = Image.new("RGB", (64, 64), (220, 30, 30))
	# Four predecoded RGB frames, with timestamps supplied through video_metadata.
	video = np.stack([np.asarray(Image.new("RGB", (64, 64), color)) for color in
	[(220, 30, 30), (30, 220, 30), (30, 30, 220), (220, 220, 30)]])
	messages = [{"role": "user", "content": [
	{"type": "image", "image": image},
	{"type": "video", "video": video},
	{"type": "text", "text": "Describe the image and the video."},
	]}]
	inputs = processor.apply_chat_template(
	messages, tokenize=True, add_generation_prompt=True,
	return_dict=True, return_tensors="pt",
	do_sample_frames=False,
	video_metadata=[VideoMetadata(total_num_frames=4, fps=2, frames_indices=[0, 1, 2, 3])],
	).to(model.device)
	output = model.generate(**inputs, do_sample=False, max_new_tokens=10)
	print(processor.decode(output[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True))
	```

	### Codes to create this repo (✅ verified):

	<details><summary>Click to expand</summary>

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

	import torch
	from huggingface_hub import hf_hub_download, list_repo_files
	from safetensors.torch import save_file
	from transformers import AutoConfig, AutoModelForImageTextToText, set_seed

	source_model_id = "Qwen/Qwen3.8-27B-FP8"
	save_folder = "/tmp/tiny-random/qwen38-fp8"
	source_revision = '017b9c7af6b5689d5dd426a76e0bc077eb5ca20a'
	set_seed(42)
	folder = Path(save_folder)
	folder.mkdir(parents=True, exist_ok=True)
	raw = json.loads(Path(hf_hub_download(source_model_id, "config.json", revision=source_revision)).read_text())
	quantization = raw.pop("quantization_config")
	quantization["weight_block_size"] = [32, 32]
	raw["text_config"].update(
	hidden_size=32, head_dim=128, num_hidden_layers=8,
	layer_types=(["linear_attention"] * 3 + ["full_attention"]) * 2,
	num_attention_heads=12, num_key_value_heads=2,
	linear_num_key_heads=2, linear_num_value_heads=6,
	linear_key_head_dim=128, linear_value_head_dim=128,
	intermediate_size=512,
	)
	raw["text_config"]["rope_parameters"]["mrope_section"] = [6, 5, 5]
	raw["vision_config"].update(depth=2, hidden_size=64, intermediate_size=128, num_heads=2, out_hidden_size=32)
	(folder / "config.json").write_text(json.dumps(raw, indent=2))
	config = AutoConfig.from_pretrained(folder)
	model = AutoModelForImageTextToText.from_config(config, dtype=torch.bfloat16)
	backbone = model.model.language_model
	text = backbone.config
	def norm():
	return type(backbone.norm)(text.hidden_size, text.rms_norm_eps).to(torch.bfloat16)
	model.mtp = torch.nn.ModuleDict({
	"pre_fc_norm_embedding": norm(),
	"pre_fc_norm_hidden": norm(),
	"fc": torch.nn.Linear(2 * text.hidden_size, text.hidden_size, bias=False, dtype=torch.bfloat16),
	"layers": torch.nn.ModuleList([deepcopy(backbone.layers[-1])]),
	"norm": norm(),
	})
	with torch.no_grad():
	for name, param in sorted(model.named_parameters()):
	if name.endswith("A_log"):
	param.zero_()
	elif name.endswith("dt_bias"):
	param.fill_(1)
	elif "norm" in name and name.endswith("weight"):
	param.fill_(1 if "visual" in name or ".linear_attn.norm." in name else 0)
	elif name.endswith("bias"):
	param.zero_()
	else:
	torch.nn.init.normal_(param, mean=0, std=0.02)
	state = {name: value.detach().cpu().contiguous() for name, value in model.state_dict().items()}
	# Prune source exclusions absent from this architecture before FP8 loading.
	names = set(dict(model.named_modules())) \| set(dict(model.named_parameters()))
	exclusions = [name for name in quantization["modules_to_not_convert"] if name in names]
	quantization["modules_to_not_convert"] = exclusions

	def pack(name, weight):
	n, k = weight.shape
	bn, bk = quantization["weight_block_size"]
	blocks = weight.float().reshape(n // bn, bn, k // bk, bk)
	scales = (blocks.abs().amax(dim=(1, 3)) / 448).clamp_min(1e-12).to(torch.bfloat16)
	packed = (blocks / scales.float()[:, None, :, None]).clamp(-448, 448)
	state[name] = packed.reshape(n, k).to(torch.float8_e4m3fn).contiguous()
	state[name.removesuffix("weight") + "weight_scale_inv"] = scales.contiguous()

	for name, module in model.named_modules():
	excluded = any(name == prefix or name.startswith(prefix + ".") for prefix in exclusions)
	if isinstance(module, torch.nn.Linear) and not excluded:
	pack(name + ".weight", state[name + ".weight"])
	save_file(state, folder / "model.safetensors", metadata={"format": "pt"})
	raw["quantization_config"] = quantization
	(folder / "config.json").write_text(json.dumps(raw, indent=2))
	# Preserve tokenizer, processor and generation assets exactly as published.
	assets = {
	"tokenizer.json", "tokenizer_config.json", "special_tokens_map.json",
	"vocab.json", "merges.txt", "added_tokens.json", "chat_template.jinja",
	"chat_template.json", "preprocessor_config.json", "processor_config.json",
	"video_preprocessor_config.json", "generation_config.json",
	}
	for name in sorted(set(list_repo_files(source_model_id, revision=source_revision)) & assets):
	copyfile(hf_hub_download(source_model_id, name, revision=source_revision), folder / name)
	```

	</details>

	### Printing the model:

	<details><summary>Click to expand</summary>

	```text
	Qwen3_5ForConditionalGeneration(
	(model): Qwen3_5Model(
	(visual): Qwen3_5VisionModel(
	(patch_embed): Qwen3_5VisionPatchEmbed(
	(proj): Conv3d(3, 64, kernel_size=(2, 16, 16), stride=(2, 16, 16))
	)
	(pos_embed): Embedding(2304, 64)
	(rotary_pos_emb): Qwen3_5VisionRotaryEmbedding()
	(blocks): ModuleList(
	(0-1): 2 x Qwen3_5VisionBlock(
	(norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True, bias=True)
	(norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True, bias=True)
	(attn): Qwen3_5VisionAttention(
	(qkv): Linear(in_features=64, out_features=192, bias=True)
	(proj): Linear(in_features=64, out_features=64, bias=True)
	)
	(mlp): Qwen3_5VisionMLP(
	(linear_fc1): Linear(in_features=64, out_features=128, bias=True)
	(linear_fc2): Linear(in_features=128, out_features=64, bias=True)
	(act_fn): GELUTanh()
	)
	)
	)
	(merger): Qwen3_5VisionPatchMerger(
	(norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True, bias=True)
	(linear_fc1): Linear(in_features=256, out_features=256, bias=True)
	(act_fn): GELU(approximate='none')
	(linear_fc2): Linear(in_features=256, out_features=32, bias=True)
	)
	)
	(language_model): Qwen3_5TextModel(
	(embed_tokens): Embedding(248320, 32)
	(layers): ModuleList(
	(0-2): 3 x Qwen3_5DecoderLayer(
	(linear_attn): Qwen3_5GatedDeltaNet(
	(conv1d): Conv1d(1280, 1280, kernel_size=(4,), stride=(1,), padding=(3,), groups=1280, bias=False)
	(norm): Qwen3_5RMSNormGated()
	(out_proj): Linear(in_features=768, out_features=32, bias=False)
	(in_proj_qkv): Linear(in_features=32, out_features=1280, bias=False)
	(in_proj_z): Linear(in_features=32, out_features=768, bias=False)
	(in_proj_b): Linear(in_features=32, out_features=6, bias=False)
	(in_proj_a): Linear(in_features=32, out_features=6, bias=False)
	)
	(mlp): Qwen3_5MLP(
	(gate_proj): Linear(in_features=32, out_features=512, bias=False)
	(up_proj): Linear(in_features=32, out_features=512, bias=False)
	(down_proj): Linear(in_features=512, out_features=32, bias=False)
	(act_fn): SiLUActivation()
	)
	(input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
	(post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
	)
	(3): Qwen3_5DecoderLayer(
	(self_attn): Qwen3_5Attention(
	(q_proj): Linear(in_features=32, out_features=3072, bias=False)
	(k_proj): Linear(in_features=32, out_features=256, bias=False)
	(v_proj): Linear(in_features=32, out_features=256, bias=False)
	(o_proj): Linear(in_features=1536, out_features=32, bias=False)
	(q_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
	(k_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
	)
	(mlp): Qwen3_5MLP(
	(gate_proj): Linear(in_features=32, out_features=512, bias=False)
	(up_proj): Linear(in_features=32, out_features=512, bias=False)
	(down_proj): Linear(in_features=512, out_features=32, bias=False)
	(act_fn): SiLUActivation()
	)
	(input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
	(post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
	)
	(4-6): 3 x Qwen3_5DecoderLayer(
	(linear_attn): Qwen3_5GatedDeltaNet(
	(conv1d): Conv1d(1280, 1280, kernel_size=(4,), stride=(1,), padding=(3,), groups=1280, bias=False)
	(norm): Qwen3_5RMSNormGated()
	(out_proj): Linear(in_features=768, out_features=32, bias=False)
	(in_proj_qkv): Linear(in_features=32, out_features=1280, bias=False)
	(in_proj_z): Linear(in_features=32, out_features=768, bias=False)
	(in_proj_b): Linear(in_features=32, out_features=6, bias=False)
	(in_proj_a): Linear(in_features=32, out_features=6, bias=False)
	)
	(mlp): Qwen3_5MLP(
	(gate_proj): Linear(in_features=32, out_features=512, bias=False)
	(up_proj): Linear(in_features=32, out_features=512, bias=False)
	(down_proj): Linear(in_features=512, out_features=32, bias=False)
	(act_fn): SiLUActivation()
	)
	(input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
	(post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
	)
	(7): Qwen3_5DecoderLayer(
	(self_attn): Qwen3_5Attention(
	(q_proj): Linear(in_features=32, out_features=3072, bias=False)
	(k_proj): Linear(in_features=32, out_features=256, bias=False)
	(v_proj): Linear(in_features=32, out_features=256, bias=False)
	(o_proj): Linear(in_features=1536, out_features=32, bias=False)
	(q_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
	(k_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
	)
	(mlp): Qwen3_5MLP(
	(gate_proj): Linear(in_features=32, out_features=512, bias=False)
	(up_proj): Linear(in_features=32, out_features=512, bias=False)
	(down_proj): Linear(in_features=512, out_features=32, bias=False)
	(act_fn): SiLUActivation()
	)
	(input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
	(post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
	)
	)
	(norm): Qwen3_5RMSNorm((32,), eps=1e-06)
	(rotary_emb): Qwen3_5TextRotaryEmbedding()
	)
	)
	(lm_head): Linear(in_features=32, out_features=248320, bias=False)
	(mtp): ModuleDict(
	(pre_fc_norm_embedding): Qwen3_5RMSNorm((32,), eps=1e-06)
	(pre_fc_norm_hidden): Qwen3_5RMSNorm((32,), eps=1e-06)
	(fc): Linear(in_features=64, out_features=32, bias=False)
	(layers): ModuleList(
	(0): Qwen3_5DecoderLayer(
	(self_attn): Qwen3_5Attention(
	(q_proj): Linear(in_features=32, out_features=3072, bias=False)
	(k_proj): Linear(in_features=32, out_features=256, bias=False)
	(v_proj): Linear(in_features=32, out_features=256, bias=False)
	(o_proj): Linear(in_features=1536, out_features=32, bias=False)
	(q_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
	(k_norm): Qwen3_5RMSNorm((128,), eps=1e-06)
	)
	(mlp): Qwen3_5MLP(
	(gate_proj): Linear(in_features=32, out_features=512, bias=False)
	(up_proj): Linear(in_features=32, out_features=512, bias=False)
	(down_proj): Linear(in_features=512, out_features=32, bias=False)
	(act_fn): SiLUActivation()
	)
	(input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
	(post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06)
	)
	)
	(norm): Qwen3_5RMSNorm((32,), eps=1e-06)
	)
	)
	```

	</details>

	### Test environment:

	- Pillow: 12.2.0
	- accelerate: 1.15.0
	- av: 19.0.2
	- flashinfer-python: 0.7.2
	- huggingface_hub: 1.33.0
	- kernels: 0.17.0
	- numpy: 2.3.5
	- safetensors: 0.8.0
	- sglang: 0.5.23.dev384+g9ab74b851
	- torch: 2.14.1
	- torchcodec: 0.17.0
	- transformers: 5.19.0

	### Change log:

	- 2026-10-23: Initial version.
	- 2026-10-23: Reduced all repo files below 50 MB; GPU-validated TP=2, MTP and CUDA Graph.
	- 2026-10-23: Creation examples use only public dependencies; source assets are copied unchanged.

	This repo is generated by GPT-6.1-Sol