Instructions to use vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt with Transformers:

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

pipe = pipeline("text-generation", model="vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt")
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("vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt")
model = AutoModelForMultimodalLM.from_pretrained("vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt", 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 vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt

SGLang

How to use vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt 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 "vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt" \
    --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": "vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt",
		"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 "vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt" \
        --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": "vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt with Docker Model Runner:
```
docker model run hf.co/vroomfondel/Qwen3.8-27B-NVFP4-ModelOpt
```

vroomfondel commited on Aug 15

Commit

3d35db5

verified ·

1 Parent(s): fd02804

Upload NVFP4 (ModelOpt) quantization

Browse files

Files changed (1) hide show

README.md +73 -4

README.md CHANGED Viewed

@@ -1,18 +1,87 @@
 ---
 library_name: transformers
 license: apache-2.0
 base_model: Qwen/Qwen3.8-27B
 base_model_relation: quantized
 pipeline_tag: text-generation
 tags:
-- modelopt
 - nvfp4
 - fp4
 ---
-# qwen3.8-27b-nvfp4-modelopt
-NVFP4 quantization of [Qwen/Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B) via NVIDIA ModelOpt.
 <!-- BEGIN quant:preview -->

 ---
 library_name: transformers
 license: apache-2.0
+license_link: https://huggingface.co/Qwen/Qwen3.8-27B/blob/main/LICENSE
 base_model: Qwen/Qwen3.8-27B
 base_model_relation: quantized
 pipeline_tag: text-generation
 tags:
 - nvfp4
+- modelopt
 - fp4
+- dense
+- multimodal
+- gated-deltanet
+- hybrid
+- mtp
+- speculative-decoding
 ---
+# Qwen3.8-27B NVFP4 (ModelOpt, uniform W4A4)
+NVFP4 (W4A4) quantization of [Qwen/Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B),
+produced with the NVIDIA TensorRT Model Optimizer (ModelOpt). This is a derivative
+checkpoint; see the [base model card](https://huggingface.co/Qwen/Qwen3.8-27B) for the
+model's capabilities and general usage.
+| | |
+|---|---|
+| **Base model** | [Qwen/Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B) (Qwen3.8 dense, multimodal, apache-2.0) |
+| **Architecture** | `Qwen3_5ForConditionalGeneration`, Gated-DeltaNet hybrid (64 layers: 48 linear-attention, 16 full-attention), dense FFN, 27B |
+| **Precision** | uniform W4A4 NVFP4 across dense FFN, full attention and the Gated-DeltaNet path; FP8 KV cache with baked scales. Vision tower, embeddings, `lm_head` and the NEXTN/MTP head kept BF16. |
+| **Size** | 55.6 GB BF16 source, 20.6 GB quantized (2.7x): 12.2 GB packed NVFP4 weights, 6.9 GB BF16 (embeddings, `lm_head`, vision, MTP), 1.5 GB FP8 block scales |
+| **Context** | 262144 native (1M needs YaRN RoPE scaling, not configured here) |
+| **GSM8K** | 97.42% (1285/1319), zero errors, zero empty responses. See the benchmark table. |
+| **Producer** | NVIDIA ModelOpt via the kikube single-node quant driver |
+> **Serving note:** this is a *uniform* W4A4 build, i.e. attention is quantized too.
+> NVIDIA's own Qwen3.5/3.6 NVFP4 checkpoints are MoE/FFN-only and leave attention in
+> BF16, so SGLang's stock `qwen3_5` path has never had to handle this. See the
+> requirements below before serving.
+### Speculative decoding
+Use **DSPARK** with the external drafter
+[RadixArk/Qwen3.8-27B-DSpark](https://huggingface.co/RadixArk/Qwen3.8-27B-DSpark)
+(1.36B BF16, 2.6 GB). **NEXTN on this checkpoint's built-in MTP head crashes the SGLang
+scheduler** on the first request (FlashInfer `fast_prefill_plan`); DSPARK does not.
+```
+--speculative-algo DSPARK \
+--speculative-draft-model-path <drafter> \
+--speculative-draft-model-quantization unquant \
+--speculative-dspark-block-size 7 \
+--speculative-num-draft-tokens 8 \
+--speculative-num-steps 1 \
+--speculative-eagle-topk 1 \
+--mamba-scheduler-strategy extra_buffer \
+--attention-backend flashinfer
+```
+Three of these are not optional:
+- `--speculative-draft-model-quantization unquant` — SGLang otherwise inherits this
+  target's `modelopt_fp4` for the BF16 drafter and the load dies on a state-dict shape
+  mismatch.
+- `--speculative-num-draft-tokens 8` — must equal `dspark_block_size + 1`; SGLang asserts
+  it. The drafter's `config.json` declares `block_size: 7`.
+- `--mamba-scheduler-strategy extra_buffer` — required for correct Gated-DeltaNet state
+  rollback when draft tokens are rejected.
+`--attention-backend triton` is not a substitute for `flashinfer` here: it hits a
+forward-time crash in `RadixLinearAttention.forward` for this architecture
+([sgl-project/sglang#29577](https://github.com/sgl-project/sglang/issues/29577)).
+DSPARK pins `num_steps` and `eagle_topk` to 1 itself. `SGLANG_ENABLE_SPEC_V2` is a no-op on
+SGLang 0.5.17 (the V2 worker always runs); the corresponding knob is
+`--disable-overlap-schedule`.
+**Measured throughput** on one DGX Spark (GB10, sm121), SGLang 0.5.17, over 415 decode
+batches during the GSM8K run: **~186 tok/s aggregate** at 11-13 concurrent requests
+(roughly 15 tok/s per stream), mean accept len **3.77** of a ceiling of 8, accept rate
+0.38-0.42. No same-settings run without speculation was measured, so this is not a speedup
+factor. Note the Mamba state cache, not the GPU, is what caps concurrency here: state usage
+sat at 0.68-0.80 with requests queued, and speculation consumes more state slots than plain
+decoding.
 <!-- BEGIN quant:preview -->