Instructions to use tiny-random/mistral-large-4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tiny-random/mistral-large-4 with Transformers:

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

pipe = pipeline("image-text-to-text", model="tiny-random/mistral-large-4")
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/mistral-large-4")
model = AutoModelForMultimodalLM.from_pretrained("tiny-random/mistral-large-4", 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/mistral-large-4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tiny-random/mistral-large-4"
# 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/mistral-large-4",
		"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/mistral-large-4

SGLang

How to use tiny-random/mistral-large-4 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/mistral-large-4" \
    --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/mistral-large-4",
		"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/mistral-large-4" \
        --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/mistral-large-4",
		"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/mistral-large-4 with Docker Model Runner:
```
docker model run hf.co/tiny-random/mistral-large-4
```

yujiepan commited on 1 day ago

Commit

0434c2c

verified ·

1 Parent(s): cc8624c

Upload folder using huggingface_hub

Browse files

Files changed (4) hide show

.meta/create_model.py +2 -3
README.md +3 -3
config.json +0 -1
consolidated.safetensors +2 -2

.meta/create_model.py CHANGED Viewed

@@ -60,7 +60,6 @@ def create_model(source_model_id: str, source_revision: str, save_folder: str):
         "quant_method": "fp8",
         "activation_scheme": "dynamic",
         "weight_block_size": [64, 64],
-        "scale_fmt": "ue8m0",
         "modules_to_not_convert": [
             "model.vision_tower",
             "model.multi_modal_projector",
@@ -199,7 +198,7 @@ def create_model(source_model_id: str, source_revision: str, save_folder: str):
         rows, columns = value.shape
         blocks = value.float().reshape(rows // 64, 64, columns // 64, 64)
         maximum = blocks.abs().amax(dim=(1, 3)).clamp_min(torch.finfo(torch.float32).tiny)
-        inverse = torch.pow(2.0, torch.ceil(torch.log2(1.0 / maximum))).to(torch.float8_e8m0fnu)
         state[name] = (
             (blocks * inverse.float()[:, None, :, None])
             .clamp(-448, 448)
@@ -207,7 +206,7 @@ def create_model(source_model_id: str, source_revision: str, save_folder: str):
             .to(torch.float8_e4m3fn)
             .contiguous()
         )
-        state[name.removesuffix(".weight") + ".qscale_weight"] = (1.0 / inverse.float()).to(torch.float8_e8m0fnu)
     for name, value in list(state.items()):
         if should_quantize(name, value):

         "quant_method": "fp8",
         "activation_scheme": "dynamic",
         "weight_block_size": [64, 64],
         "modules_to_not_convert": [
             "model.vision_tower",
             "model.multi_modal_projector",
         rows, columns = value.shape
         blocks = value.float().reshape(rows // 64, 64, columns // 64, 64)
         maximum = blocks.abs().amax(dim=(1, 3)).clamp_min(torch.finfo(torch.float32).tiny)
+        inverse = torch.pow(2.0, torch.ceil(torch.log2(1.0 / maximum)))
         state[name] = (
             (blocks * inverse.float()[:, None, :, None])
             .clamp(-448, 448)
             .to(torch.float8_e4m3fn)
             .contiguous()
         )
+        state[name.removesuffix(".weight") + ".qscale_weight"] = (1.0 / inverse).contiguous()
     for name, value in list(state.items()):
         if should_quantize(name, value):

README.md CHANGED Viewed

@@ -9,8 +9,8 @@ This is a randomly initialized tiny model based on [mistralai/Mistral-Large-4-1T
 | Feature | Source | This repo |
 |---|---|---|
-| Checkpoint tensor elements | 7.13B BF16, 63.84M F32, 1.05T F8_E4M3 | 54.84M BF16, 1.31M F8_E4M3, 320 F8_E8M0 |
-| Safetensors weight files | 1.06 TB (54 files) | 111.12 MB (consolidated.safetensors) |
 | Architecture / input | Multimodal MoE; text and image | Same |
 | Vocabulary rows / tied embeddings | 204,800 / false | 204,800 / true |
 | Backbone layers | 72 | 12: three complete RoPE/RoPE/RoPE/NoPE cycles |
@@ -20,7 +20,7 @@ This is a randomly initialized tiny model based on [mistralai/Mistral-Large-4-1T
 | Cross-layer KV sharing | Source pattern | Two complete three-own/one-shared NoPE cycles |
 | Vision tower | 30 layers; 2,048 width | 4 layers; 64 width |
 | DSpark | 3 dense MLA stages; rank 512; depth 3 | 3 dense MLA stages; rank 8; depth 3 |
-| Checkpoint precision | E4M3 block-FP8 + UE8M0 scales; BF16 exclusions | Same format; 64×64 blocks |
 TP=2 vLLM uses a 64-wide FP8 activation group per rank, which fixes the residual width at 128. The checkpoint retains 64×64 FP8 blocks, a 512-wide MLA latent KV required by H20 sparse MLA, and four 32-wide attention heads. The input/output BF16 token table is tied, saving one 52.43 MB matrix; the optimized vLLM kernel constraints still place the checkpoint above the 30 MB soft target.

 | Feature | Source | This repo |
 |---|---|---|
+| Checkpoint tensor elements | 7.13B BF16, 63.84M F32, 1.05T F8_E4M3 | 54.84M BF16, 320 F32, 1.31M F8_E4M3 |
+| Safetensors weight files | 1.06 TB (54 files) | 111.13 MB (consolidated.safetensors) |
 | Architecture / input | Multimodal MoE; text and image | Same |
 | Vocabulary rows / tied embeddings | 204,800 / false | 204,800 / true |
 | Backbone layers | 72 | 12: three complete RoPE/RoPE/RoPE/NoPE cycles |
 | Cross-layer KV sharing | Source pattern | Two complete three-own/one-shared NoPE cycles |
 | Vision tower | 30 layers; 2,048 width | 4 layers; 64 width |
 | DSpark | 3 dense MLA stages; rank 512; depth 3 | 3 dense MLA stages; rank 8; depth 3 |
+| Checkpoint precision | E4M3 block-FP8 + FP32 scales; BF16 exclusions | Same format; 64×64 blocks |
 TP=2 vLLM uses a 64-wide FP8 activation group per rank, which fixes the residual width at 128. The checkpoint retains 64×64 FP8 blocks, a 512-wide MLA latent KV required by H20 sparse MLA, and four 32-wide attention heads. The input/output BF16 token table is tied, saving one 52.43 MB matrix; the optimized vLLM kernel constraints still place the checkpoint above the 30 MB soft target.

config.json CHANGED Viewed

@@ -14,7 +14,6 @@
       64,
       64
     ],
-    "scale_fmt": "ue8m0",
     "modules_to_not_convert": [
       "model.vision_tower",
       "model.multi_modal_projector",

       64,
       64
     ],
     "modules_to_not_convert": [
       "model.vision_tower",
       "model.multi_modal_projector",

consolidated.safetensors CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:233e1a76d998d19bf117cf6acc381c31eeb46a3313811543d60f7d42cfaada56
-size 111124456

 version https://git-lfs.github.com/spec/v1
+oid sha256:ab784992be89268fd19a096a69a31216896129dd6703af7339411b09d7e4bf4a
+size 111125104